Make the http timeout test deterministic with a fake clock (#324)

[LinuxDevil]

Closes #324

What made it flaky

src/tools/built-in/http.test.ts "rejects near the configured timeout" ran a real request against a real server that never answers and asserted 450 <= elapsed < 2000 against the real 500 ms timeout. Under load (the full coverage run, many vitest workers) the event loop can be delayed by more than 1.5 s, so the upper bound failed. Production code is fine: the timer is an ordinary setTimeout(() => controller.abort(), timeoutMs) in runHttpRequest.

Fix (test only, no production change)

The test now fakes only setTimeout/clearTimeout (vi.useFakeTimers({ toFake: [...] })); the sockets stay real. It waits until the server has the request (so the timer is armed), then asserts: still pending after 499 ms, rejects with "timed out after 500ms" at 500 ms, and the server sees the connection close (the abort tore it down). That is stricter than before (exact boundary and the exact message, plus the connection teardown) and has no wall-clock comparison. Real timers are restored in finally.

Other tests named in the ticket

credentialBroker.test.ts has no timing assertion. src/execution/guardrails.test.ts has elapsed < 2000 checks around real child processes (100 ms command timeout); a fake clock does not apply there (the child and the kill are real), so I left them alone as the ticket says "only if the fix is the same".

Repeat-run evidence (Windows 11, Node 26)

npx vitest run src/tools src/security src/execution/guardrails.test.ts run 20 times, 4 runs at a time in parallel (about 4x the normal load): http.test.ts (23 tests) passed 20/20. Two unrelated tests failed in some of those 20 runs, not touched here: src/security/SubprocessSandbox.test.ts integration (Docker daemon required) in 5 runs and src/tools/workspace/checkpoints.node.test.ts "rewinds three turns" in 1 run.

Verification

• npx tsc --noEmit: clean
• npm run lint: 0 warnings
• npm run test:coverage (after merging origin/main at fe54100): 268 files passed, 1 skipped; 3878 tests passed, 6 skipped
• npm run fallow: exit 0
• npm run docs:llms:check: clean
• Not run (tests-only change, nothing in build, exports, docs or Agent Forge touched): docs snippets, test:types, pack-smoke, Agent Forge checks.

🤖 Generated with Claude Code