[gouyt13]

Summary

Changes to test.yaml previously skipped its ShellCheck job and Python summary-verdict tests. Select both checks when that workflow changes.

Benchmark source changes previously selected broad CI without compiling benchmark targets. Add an Ubuntu Benchmark build and smoke job with benchmarks ON, ordinary tests/samples OFF, native optimization OFF, and bindings ON for the Python ID helper. Build the targets and run their 15 existing small CTest cases with a per-test timeout; no external dataset, timing threshold or performance comparison.

Wire a dedicated benchmarks group through change detection. Core/hotpath benchmark-only edits select smoke plus applicable formatting/lint rather than unrelated wheel/platform matrices; shared source, binding and build/package changes also select smoke. Unknown-input, release and detection-failure safeguards remain conservative. Include hotpath Python helpers in Ruff and update coverage documentation.

Separate agent workflow templates are excluded. Required-check registration for the new job remains a follow-up after GitHub validation.

Testing

• Routing regression failed in 13 cases on the original classifier; 27 helper tests and 239 subtests now pass.
• Isolated portable Release build succeeded with GCC 14.3 and the existing Python 3.13 environment; all 15 CTest smoke cases passed, including Python IDs. No package was installed or replaced.
• Project Python checks, focused Ruff, ShellCheck, YAML output/job wiring checks, strict MkDocs, local links and diff checks passed.
• Independent read-only review passed and independently reran 27 helper unittest cases.

Cloud Ubuntu/Python 3.11 validation is pending this PR. Local build emitted existing vendored hnswlib unused-function warnings, retained without suppression. No all-ISA coverage or performance improvement is claimed. This PR intentionally selects full CI because it changes the classifier.