test: fix flaky CI in flock stress + run-process subprocesses

[filipeforattini]

Two unrelated flakes tripping CI on the 0.8.0 release:

1. flock stress flake (Windows only)

packages/core/test/util/effect-flock.test.ts:330 "enforces mutual exclusion under process contention" failed on Windows with one of 16 workers exiting 1.

Root cause: the active marker file lives outside the lock directory, so its wx create races between a holder's fs.rm(active) and the next holder's fs.writeFile(active). On Windows the race window is wide enough that one worker keeps losing the race even though the flock itself is correct.

Fix: drop the active marker from this specific test. The serialized work and done.log line count below already prove mutual exclusion. The active mechanism stays in place for the single-worker "recovers after a crashed lock owner" test where there's no race.

2. run-process flake (Linux CI, all 0.8.0 candidate tests)

packages/opencode/test/cli/run/run-process.test.ts failed three times on Linux at exactly 30s — appProc.run was timing out.

Root cause: the harness spawned bun run --conditions=browser src/index.ts per subprocess. That's ~20s of transpile + plugin init per spawn. With 11 cliIt.concurrent tests starting at once on a shared-CPU CI runner, individual startup times stretched past the 30s timeoutMs baked into the harness.

Fix: prefer the prebuilt redcode binary (dist/redcode-linux-x64/bin/redcode) when present; fall back to bun otherwise. Local run of run-process.test.ts dropped from ~60s to ~40s. CI should be well under the 30s timeoutMs per test.

Verification

• bun test test/util/effect-flock.test.ts — 11 pass, 0 fail
• bun test test/cli/run/run-process.test.ts — 13 pass, 0 fail, 3 consecutive runs

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}