feat(skills): timers-modal-and-threading skill, snippet and example

[TMHSDigital]

Closes #381. (Rebased onto main after #445 merged.)

What ships

• Skill timers-modal-and-threading:
  • bpy.app.timers return values and persistent=True;
  • modal operators driven by event_timer_add;
  • worker threads that hand results to the main thread through a queue.Queue;
  • why none of it runs in --background.
• Snippet snippets/thread-queue-timer.py.
• Check-only example examples/timers-modal-threading:
  • runs a --background child, and a windowed child that quits itself from a timer and reports JSON;
  • falsifiers --return-zero-once (exit 5) and --non-persistent (exit 8).
• Wiring: catalog row; examples/skills.json (→ this skill and operators); Cursor manifest; counts (18 skills, 29 snippets, 66 examples, 10 check-only).

Measured facts the skill relies on (4.5.11, 5.1.2, 5.2.1)

Fact	Result
Timer in a --background child	is_registered True, never runs
None timer / timer returning 0.05 until its 3rd call	runs 1× / 3×, then unregistered
Run-once timer returning 0.0	re-runs (5× on 5.2.1 before a file load removed it)
register(fn, 0.5)	TypeError: register() takes exactly 1 positional argument (2 given) on all three. 5.2's docstring shows the *; 4.5's doesn't, but the behaviour is the same
wm.read_factory_settings from a timer	the persistent timer stays registered, the plain one is dropped
Queue drained by a timer	runs on the main thread; mesh created from the worker's result
Modal operator via INVOKE_DEFAULT under temp_override(window=…)	RUNNING_MODAL, 3 TIMER events, FINISHED, timer removed
'timer' in bpy.types.Event.bl_rna.properties	False (4.5.11 and 5.2.1). event.timer raised AttributeError in a 5.2.1 modal, which first hung my probe

Found while building: the first windowed run hung, because the timer meant to quit Blender was itself dropped by the file load the example performs. That is the contract under test. The quit timer is now persistent=True, and the skill says so.

Evidence

Live run: tests/smoke/run_catalog.py over the new row on local Windows binaries; all PASS.

Binary reports	Happy path	--return-zero-once	--non-persistent
Blender 4.5.11 LTS	PASS	PASS (exit 5)	PASS (exit 8)
Blender 5.1.2	PASS	PASS (exit 5)	PASS (exit 8)
Blender 5.2.1 LTS	PASS	PASS (exit 5)	PASS (exit 8)

Local checks: tests/run_all.py 31 passed, 0 failed.

Not yet proven: the windowed child on Linux under xvfb. This PR's smoke legs are the first such run. If the child cannot open a window there, the row fails with exit 4 rather than passing silently.

🤖 Generated with Claude Code

[TMHSDigital]

CI evidence before merge: all checks pass ( 13 pass ). Blender Smoke on PR head: Blender 5.2.2 LTS / Blender 4.5.14 LTS. Summaries (happy path; falsifiers) per leg: 145 passed, 1 skipped, 0 failed;166 passed, 0 skipped, 0 failed;143 passed, 3 skipped, 0 failed;151 passed, 15 skipped, 0 failed;

[TMHSDigital]

Linux evidence (run 37258749968, head 525191b): the windowed child ran under xvfb, [PASS] timers-modal-threading plus both falsifiers. Under --return-zero-once the run-once timer ran 4x on Linux (5x locally on Windows); any count other than 1 trips exit 5, so the timing difference does not affect the check.