tests: #807 — async-context tracking harness (ALS + async_hooks)

[proggeramlug]

Summary

New test-files/test_harness_async_context.ts covers the propagation rules AsyncLocalStorage + async_hooks need to follow, with eight independent sections each printing section: result:

1. Sync run() store baseline.
2. Single await to a separately-named async fn.
3. Chained awaits (depth 8).
4. queueMicrotask callback inside run().
5. setTimeout(fn, 0) callback inside run().
6. Nested run() with an await in each layer.
7. Concurrent Promise.all of two runs (isolation, no cross-pollination).
8. createHook lifecycle + executionAsyncId API shape.

What it surfaces today

Verified against node --experimental-strip-types (16 stable lines, all sections PASS). Perry v0.5.912 has partial support:

• Pass: sections 1, 6, 7. Sync nested runs propagate correctly when getStore() is called inline in the run body, and concurrent isolation holds.
• Fail: sections 2/3 (await to a separately-named async fn) and 4/5 (queueMicrotask, setTimeout) — store evaporates after the suspension; getStore() returns undefined.
• Stub: section 8 — createHook returns undefined; API is name-only today.

That partial picture is itself the useful signal — the gap isn't all-or-nothing.

known_failures.json

Adds test_harness_async_context linked to #788 (AsyncLocalStorage real tracking) and #789 (async_hooks createHook lifecycle). Each section flips to PASS independently as those land.

Test plan

• node --experimental-strip-types produces 16 stable lines.
• perry compile && /tmp/out runs end-to-end, exits 0 (it doesn't crash — it just prints undefined for the failing sections), and the diff pins exactly which propagation rules are missing.
• known_failures.json entry prevents the parity gate from flagging this as a new regression.

Closes #807.