refactor: introduce app_common static library for app-layer glue

[hsheng27120]

Description

libslic3r_gui's symbol closure depended on app-layer glue (Sentry wrapper, bury-point state, BaseException/StackWalker) that only the production executable compiled, so downstream consumers (tests, headless tools) needed stub files or the Snapmaker_Orca import library to link against the GUI library.

Root cause: N/A (build/layout refactor, not a bug fix).

Fix:

• New src/app_common/ static library: moves SentryWrapper (from src/sentry_wrapper/), bury_point (from src/bury_cfg/), and BaseException/StackWalker (from src/dev-utils/) under one target; common_func.cpp is compiled there too because SentryWrapper.obj references common:: symbols needed by libslic3r_gui's closure.
• Phase-1 shadow library: the production targets (Snapmaker_Orca DLL/exe and the Windows shim) still compile the same sources directly, so runtime behavior is unchanged; the new library initially serves libslic3r_gui and its downstream only.
• Three-state BURY_API (import-from-dll / dllexport / plain) so references no longer default to dllimport and require an import library. The shim keeps its existing BURY_EXPORTS local bury-point copy (see the MESA/opengl32 loader note in src/CMakeLists.txt), so its runtime design is untouched.
• ~25 GUI/desktop files updated to the new app_common/... include path; app_common linked into libslic3r_gui.

Why this is safe / Compatibility:

• The shim still compiles bury_point.cpp locally with BURY_EXPORTS and does not statically depend on Snapmaker_Orca.dll — the MESA software-OpenGL fallback design is preserved.
• With the default (plain) BURY_API, references bind to same-image definitions, which is semantically equivalent to the previous dllimport usage within each image.
• Removed macros (BP_LOCAL_SERVER_ERR_CODE and a duplicate BP_LOCAL_SERVER define) have no remaining users; two stale comments about include resolution were corrected.
• Snapmaker_Orca_app_gui, Snapmaker_Orca DLL, and libslic3r_gui link and resolve cleanly with both the direct glue objects and the app_common archive coexisting (no LNK2005 duplicate pulls).

Behavior changes: None. Source moves + build graph only; no runtime code path changes (phase-1 shadow library).

Known limits: None.

Reviewer focus:

• src/CMakeLists.txt: the app_common target definition, the shim section (kept upstream's local-bury design), and target_link_libraries(libslic3r_gui app_common) (plain signature, matching this target's existing style).
• Include-path resolution: quoted includes relative to src/-rooted headers plus app_common's PUBLIC include dir propagating through libslic3r_gui.

Screenshots/Recordings/Graphs

N/A (no UI change)

Tests

• Build: Windows x64, VS 2022 (MSVC 19.44), Release — PASS. Verified targets: configure (GUI=ON, SENTRY=ON), app_common, libslic3r_gui (incremental rebuild of the 25 include-updated files), Snapmaker_Orca.dll link (no duplicate symbols), snapmaker-orca.exe shim link (BURY_EXPORTS path). CMake configure also verified clean on this PR branch against build_platform_engineering.
• Automated tests: libslic3r_tests [SSWCPProtocol] — PASS (77 assertions in 10 test cases), headless (SLIC3R_GUI=OFF) configuration.
• CI: NOT AVAILABLE (expected to run on this PR).
• Manual verification:
  1. Configure + build with default options (SLIC3R_GUI=ON) — Expected: configures and links all targets; Actual: PASS on Windows.
• Not verified: Linux/macOS build configurations (no environment available; CI will cover); GUI runtime launch and Sentry/bury runtime behavior (no runtime code changes intended; phase-1 keeps production compile units identical).

[zhangzhend0ng]

Review verdict: Request changes

The build-graph refactor is sound overall — direction, symbol-closure semantics, and the compatibility claims all check out on inspection. Two items to address before merge (details inline):

1. [Must fix] BURY_API's BURY_IMPORT_FROM_DLL state is defined nowhere in the tree, and its comment misattributes usage to the Windows shim (which actually compiles bury_point.cpp locally with BURY_EXPORTS). Remove the dead state or fix the comment. → inline on src/app_common/bury_point.hpp
2. [Project standard] New comments in src/app_common/bury_point.hpp, the app_common block of src/CMakeLists.txt, and the link block of src/slic3r/CMakeLists.txt are in Chinese; the project coding standard requires English. Please translate.

Verified clean

• Macro removals safe: BP_LOCAL_SERVER was defined twice in the old header; this PR removes only the duplicate. The kept define (app_common/bury_point.hpp:56) still serves the 4 uses in HttpServer.cpp; BP_LOCAL_SERVER_ERR_CODE has zero remaining users.
• No stale references: no remaining sentry_wrapper/, bury_cfg/, dev-utils/BaseException|StackWalker includes anywhere (src/tests/sandboxes); include casing is exact — the stackwalker.h → StackWalker.h change fixes a latent case-sensitivity trap.
• Target scoping: app_common and libslic3r_gui are both behind SLIC3R_GUI; encoding_check is defined before first use; tests/slic3rutils links libslic3r_gui under a TARGET libslic3r_gui guard, so usage requirements propagate to exactly the consumers this PR unblocks.
• Plain-signature link is correct: libslic3r_gui's existing target_link_libraries calls are all plain; CMake forbids mixing signatures on one target, so the keyword-less target_link_libraries(libslic3r_gui app_common) is the required form here.
• Linkage equivalence: DLL keeps its dllexport bury/sentry copy (state host), shim keeps its local copy (MESA/opengl32 fallback design untouched), non-Windows exe unchanged (extern). Plain references binding to same-image definitions is semantically equivalent to the previous per-image dllimport usage; lazy archive extraction avoids duplicate symbols — author-verified on MSVC Release, and GNU ld/ld64 share the same lazy semantics.

Merge gates

• Linux/macOS builds not yet exercised (explicitly declared in the description — thanks). CI must be green on all platforms before merge.
• Advisory inline: add a phase-2 TODO for retiring the shadow-library double compilation.
• For the record, no action needed: branch name app-common-static-lib lacks the refactor_ prefix per workflow conventions.

[zhangzhend0ng]

[Must fix] BURY_IMPORT_FROM_DLL is dead state and the comment above it is factually wrong.

Grepping the entire tree at this PR head: nothing defines BURY_IMPORT_FROM_DLL — no CMake target, no source file. Meanwhile the comment claims this state is for "仅 Windows exe shim" whose bury state lives in Snapmaker_Orca.dll. But the shim actually compiles app_common/bury_point.cpp locally with BURY_EXPORTS (see src/CMakeLists.txt, shim target) and keeps its own bury state — exactly the MESA/opengl32 design the PR description says is preserved. So today the dllimport branch is unreachable, and the comment misdescribes the shim's linkage.

Please either:

• drop BURY_IMPORT_FROM_DLL and collapse to two states (BURY_EXPORTS → dllexport, default → plain), rewriting the comment; or
• if this state is reserved for a planned future consumer, name that consumer in the comment instead of the shim.

This is linkage-critical machinery — a wrong "who imports what from where" comment is exactly what breaks the next refactor of the shim.

Same block, project standard: the new comments here (and in the app_common block of src/CMakeLists.txt, and the link block of src/slic3r/CMakeLists.txt) are in Chinese. The project coding standard requires no Chinese in source code — please translate to English, which also matches the English style of the surrounding linkage comments.

[zhangzhend0ng]

[Advisory] Phase-1 shadow library: the no-duplicate-symbol outcome depends on link order — please add a phase-2 TODO.

Production targets compile the glue sources directly and link app_common (transitively via libslic3r_gui). This resolves without LNK2005 only because direct objects precede archives on the link line and archive members are extracted lazily. That invariant is implicit — a future --whole-archive, link-line reorder, or source-set drift could surface duplicate-symbol errors.

Suggestion: add a # TODO(phase-2): production targets consume app_common and stop compiling these sources directly next to this target (plus a tracking issue), so the transitional state has an exit.

Minor: target_include_directories(app_common PUBLIC ${CMAKE_CURRENT_SOURCE_DIR}) exposes all of src/ to every downstream consumer. Acceptable since it matches existing project practice; a follow-up could narrow it.