Fix bug 94: Add TTL to claims so interrupted runs don't leave stale claims

[joewalker]

Summary

This fixes bug #94 where an interrupted loop() run leaves a non-expiring claim, causing the prompt to be silently skipped forever on resume.

Root Cause

When loop() is terminated ungracefully (Ctrl+C/SIGINT/SIGKILL) while a prompt is claimed but not yet completed:

1. The claim is persisted to the state file with a runId
2. The finally block doesn't execute, so release() never runs
3. The claim remains in the persisted state, owned by a now-dead runId
4. On subsequent runs, claim() sees a claim with a different runId and returns false, causing the prompt to be skipped with "Skip (claimed elsewhere)"
5. The expiresAt field existed but was never set or checked, so the stale claim persists indefinitely

Solution

• Add CLAIM_TTL_MS constant (1 hour) for claim expiration - safely longer than any prompt's maximum runtime
• Set expiresAt when creating new claims
• Check for expired claims in claim() method and clean them up before checking ownership
• Handle legacy claims without expiresAt that are owned by a different runId by treating them as stale

Changes

• src/loop-states/file.ts: Modified the claim() method to check for expired claims and handle legacy claims

Testing

The fix ensures that:

1. New claims have an expiration time (1 hour from creation)
2. Expired claims are automatically cleaned up when claim() is called
3. Legacy claims without expiresAt that are owned by a different run are treated as stale and removed
4. The prompt can be reprocessed on subsequent runs after the claim expires or is cleaned up

Closes #94