Close playlist file handle and database connection after use

[JavaGT]

60-second summary

Two leaked resources are now closed: the playlist M3U8 file handle and the download-database connection. Verify with git diff -w 478c3f2 e5c44fe (semantic change is +4/−4 lines; the raw diff is mostly re-indentation of main() moving under a new try). Risk is minimal; rollback is git revert e5c44fe.

What

_update_playlist_file reads the playlist file inside a with block, and the CLI closes the Database in a finally after all uses (success or error).

Why

The playlist read at gamdl/downloader/downloader.py:119-123 left the file handle to the garbage collector, and Database created at gamdl/cli/cli.py:130-135 was never closed — Database.close (gamdl/cli/database.py:44-45) had no callers. Evidence and discussion: #2.

The change

• gamdl/downloader/downloader.py: with block around the playlist-file read (same data, handle now released deterministically).
• gamdl/cli/cli.py: everything after Database creation wrapped in try/finally with if database: database.close(); the database can be None on early config errors, and the close sits after the last database.add / flat_filter use.
• Intentional non-goal: the per-track yt-dlp process spawn (Perf: a fresh OS process is spawned per track for yt-dlp downloads (spawn + full yt-dlp import per song) #3) is untouched — that finding is "measure first" and stays issue-only.

Verification

• python3 -m py_compile on both touched files: clean; both modules import in the project venv.
• Synthetic probe on the real Database (temp sqlite, no network/credentials): add/get roundtrip, connection unusable after close() (sqlite3.ProgrammingError), double-close safe, flat_filter behavior, and the committed with-read shape — 5/5 PASS. Probe and output preserved at scratch-resource-hygiene/ in the workspace (the first probe run was not kept; it was re-run and saved — flagged by the test-honesty review, see below).
• No Apple Music credentials in this environment, so the end-to-end CLI/download path was not exercised.

Reviews

• Hostile review (DeepSeek v4.1 Flash): APPROVED — close ordering, None guard, and no double-close verified; one theoretical note (a raising close() could mask an in-flight exception) accepted as non-blocking.
• Test-honesty review (DeepSeek v4 Flash): MINOR — original probe artifacts weren't preserved; resolved by the coordinator re-run above.
• GPT-5.6 ("Luna") and Grok seats were unavailable this run (usage limit / HTTP 403, consistent with prior runs); DeepSeek seats covered both lenses per the route policy.

Attribution

This PR was produced with AI assistance under the direction of @JavaGT:

• Exploration, evidence verification, implementation: GLM agents (ZCode)
• Planning: GLM agents; plan audited by the consultants below
• Hostile review: DeepSeek v4.1 Flash; test-honesty review: DeepSeek v4 Flash (via OpenRouter)
  Every change was cross-reviewed and the final diff verified against the described behavior. Happy to adjust or close any part of this — tell me what doesn't fit the project's direction.

[JavaGT]

[Account-2 upstream-readiness review — 2026-09-20 · read-only adversarial verification pass; nothing in this PR, repo, or workspace was modified]

Verdict: READY-WITH-NOTES for upstream filing (one stale number to fix first). Mechanics verified at head e5c44fe (2 files, one theme; re-indentation disclosed): with-block playlist read; None-guarded DB close() after last use (downloader.py:119–123; the close sits after the final database.add :295 and flat_filter use; finally :308–310); cli.py:130–135; close() had no other callers at base. Probe real and preserved (scratch-resource-hygiene/, 5/5, re-run disclosed in the docstring). Attribution matches the run log. Base == current upstream HEAD 478c3f2 (verified — no drift).

PRE-SHIP (body-only):

1. Semantic count wrong (high): the body (and the staging log) say "+4/−4"; git diff -w 478c3f2..e5c44fe is +9/−5 (cli.py +4/−0, downloader.py +5/−5; net +4). State the real numbers.
2. Rollback phrasing (low): "rollback is git revert e5c44fe" dangles after an upstream squash — phrase it as reverting the change, not the fork SHA.
3. Wording (low): "database can be None on early config errors" — it is None when no database_path is configured; say that.

EVALUATE (high — filing mechanics): open upstream PRs glomatico#345 (hunk @ :148) and glomatico#335 (@ :285) touch cli.py inside this PR's re-indented region — either filing order needs a rebase; downloader.py has no upstream overlap; glomatico#341 sits outside the region.

Decision question for the owner: fix the count + two phrasings at pickup? (implement-it / evaluate — nothing pre-selected.)