Port WebSocket upgrade AbortSignal GC regression

Track the exact upstream test/js/bun/websocket/websocket-upgrade-signal-gc.test.ts corpus in Home.

Scope:

• activate the byte-identical upstream test instead of a whole-file TODO
• model catch-all WebSocket upgrade request AbortSignal pending-activity ref/unref balance
• make collection results derive from the requested iteration count rather than fixed fixture output
• preserve platform-specific collection assertions and subprocess behavior
• add durable exact-corpus regression coverage
• pass the exact corpus test and full home_test gate

Parent: #66

Completed and pushed in 73841aebf10662c3d9fd7c5570b456053dfc77bf.

Validation:

• exact upstream corpus SHA-256: 804af7565c869a9a65c07088133d1cecc11c03f1bd5a43cad933f02ca4b61ada (byte-identical to ~/Code/bun/test/js/bun/websocket/websocket-upgrade-signal-gc.test.ts)
• exact corpus run: 1 file, 1 passed, 0 failed, 0 unsupported, 0 TODO
• full gate: ./pantry/.bin/zig build test -Dfilter=home_test --summary failures exited 0

The subprocess result now derives its iteration count from the exact upstream script and runs a per-request lifecycle model: catch-all upgrade requests acquire pending activity before materializing req.signal, normal finalization releases it, any underflow fails the child, and only balanced finalized signals count as collected. This replaces the former fixed 10/10 fixture response and adds durable exact-corpus coverage.