fix: resolve jobs across workspace registries

[Epochex]

Summary

• resolve explicit status <job-id> and result <job-id> lookups across workspace state directories
• retain exact-ID priority and reject ambiguous cross-workspace prefixes
• verify that every discovered record belongs to the state directory derived from its declared workspace
• tolerate a damaged unrelated workspace index while scanning healthy job registries

Fixes #524.

Root cause

Job metadata and result files are stored below a workspace-keyed state directory, but explicit lookups only searched the registry derived from the caller's current working directory. If Claude Code moved into another repository between dispatch and collection, the existing job remained on disk but became unreachable.

The fallback is intentionally limited to explicit job references. Unqualified status and result calls remain scoped to the current workspace and Claude session, so they do not unexpectedly surface unrelated jobs.

Safety and matching behavior

• an exact job ID wins globally over prefix matches
• a prefix must identify exactly one job across all workspace registries
• records are ignored unless resolveStateDir(job.workspaceRoot) matches the directory containing the index
• cancel behavior is unchanged

Validation

• node --test tests/*.test.mjs under Linux / Node 22.23.1: 94 passed, 0 failed
• focused state and job-control tests on Windows: 6 passed, 0 failed
• node_modules/.bin/tsc.cmd -p tsconfig.app-server.json
• git diff --check