[ROSETTA] curiocity command evaluator has no execution timeout

[isolomatov-gd]

`src/curiocity/src/evaluators/command.ts` calls `ctx.exec(p.run, {shell:true, cwd, reject:false, all:true})` with no timeout option, so a hanging build/test/lint command (typically `gate:true`) blocks the trial's evaluate step indefinitely. The sibling `external` evaluator (`src/curiocity/src/evaluators/external.ts`) already has a configurable `timeoutSec` (default 60s) passed to execa with explicit timeout handling via `res.timedOut`. Add the same optional `timeoutSec` param to `command`'s schema and pass it through.

[github-actions]

Rosetta Triage

Classification: bug, curiocity (test harness) — confirmed against source
Priority assessment: Medium — affects CI reliability (a hung gate command blocks the whole trial/pipeline run) but has no user-facing or security impact; not urgent since it requires a misbehaving build/test/lint command to trigger.

Findings:

• Verified: src/curiocity/src/evaluators/command.ts calls ctx.exec(p.run, { shell: true, cwd, reject: false, all: true }) with no timeout option, so a hanging command runs indefinitely.
• Verified: src/curiocity/src/evaluators/external.ts already has the pattern to copy — timeoutSec (zod default 60, z.number().positive()) passed as timeout: Math.round(p.timeoutSec * 1000), with res.timedOut mapped to a thrown CuriocityError.
• command's failure model differs from external's (returns EvalResult with pass:false in a catch, vs. external throwing to produce an eval-error trial status) — a timeout in command should likely follow command's own catch-and-report convention rather than throwing, to stay consistent with its existing contract.

Suggestions:

• Add timeoutSec: z.number().positive().default(60) to commandParamsSchema, pass timeout: Math.round(p.timeoutSec * 1000) to ctx.exec, and handle res.timedOut inside the existing try block (set exitCode/details accordingly) rather than throwing, matching command's existing pass/fail reporting style.

Completeness: Issue is well-scoped and actionable as-is; no missing information needed to implement.

Automated triage by Rosetta agent