Generalize the smoke harness to per-snippet coverage + emit a coverage table

[TMHSDigital]

Context / Problem Statement

tests/smoke/run_smoke.py is excellent but exercises only ~6 hand-picked headline behaviors and blender-smoke.yml runs 4 examples. The 17 snippets and 12 skills are largely un-run in real Blender, and there is no record of what is covered. A new snippet can ship having only passed py_compile, and coverage can erode invisibly (silent-truncation risk).

Proposed Solution / Implementation Steps

1. Add tests/smoke/run_snippets.py: iterate snippets/*.py, exec each inside a reset() factory-empty scene in headless Blender, treating any raised exception as failure (mirror the try/except wrapper already in run_smoke.py).
2. Support a # smoke: skip <reason> first-line pragma for snippets that need scene state they can't self-provide; print the skipped list so skips are visible, never silent.
3. Emit a coverage table at the end: N snippets run, M skipped (reasons), K skills with a smoke check. Print it to the CI log.
4. Wire a new step into blender-smoke.yml after the existing smoke driver, on the same 5.1/4.5 matrix, xvfb-run-wrapped.

Definition of Done

• Every snippet is either executed in headless Blender or explicitly # smoke: skip-ged with a reason.
• CI log prints a run/skip/coverage summary table.
• A snippet that raises in Blender fails the job (proven once).
• Runs green on both 5.1 and 4.5 matrix legs.

---

Filed from the repository technical audit — see docs/technical-audit.md.