fix: when user remove `ai-rework`, loope read the source code in worktree and say: this is implemented

[ngthluu]

When user remove ai-rework:

1. loope read the source code in worktree, see some code are landed (since ai failed when implement half part)
2. loope thinks implemented -> not create PR and delete worktree

Actual: no PR created
Expected: loope just prompt "continue" in the same executing claude session to continue working, instead of run checker

[ngthluu]

🤖 Picked up (bug flow). Branch: ai/issue-49

[ngthluu]

🤖 Not confident enough to implement (confidence 45/100). Answer the numbered questions below in a comment, then remove the ai-needs-info label to re-queue:

The single biggest blocker: there is no existing "resume the same Claude session" mechanism anywhere in the codebase (session ids are only ever recorded for dashboard display, never replayed with --resume), so "just continue in the same session" isn't a small patch — it requires new design choices this report doesn't specify.

1. Should the fix persist a Claude session id specifically for ai-rework pickups and pass it via Resume: on the next run, a) resuming the exact prior brainstorm/debug session, b) starting a fresh session but seeding the prompt with a summary of prior progress, or c) something else?
2. Should the "already implemented" sentinel (PIPELINE_ALREADY_DONE) be suppressed entirely when picking up from ai-rework, or should it still fire but route to a different outcome (e.g., comment + reopen instead of close+delete)?
3. Does this fix apply to both the bug pipeline (no confirmation round, pipeline_bug.go:37) and the feature pipeline (has an answerer confirmation, pipeline_feature.go:76-93), or only one?
4. Should handleIssue distinguish a "rework pickup" from a "fresh eligible pickup" as a separate code path (like Continue already does for ai-stopped), and if so what marks an issue as a rework pickup vs. fresh?
5. If the daemon process restarts between the original failed run and the label removal, is there still a Claude session to resume, or should the fallback then be "start fresh but skip the already-done check"?

[ngthluu]

1. a. Also always persist claude session id, not just displaying
2. Yes, suppressed entirely ; this PIPELINE_ALREADY_DONE only trigger for fresh new ticket
3. Apply for both
4. like how Continue work
5. still a Claude session to resume

[ngthluu]

🤖 Picked up (bug flow). Branch: ai/issue-49

[ngthluu]

🤖 UAT checklist

• Remove ai-rework label from a previously parked issue → issue resumes work, no PR-less close
• Watch pickup on a rework issue → same Claude session continues instead of starting fresh
• Resumed session finds partial code already committed → it keeps implementing instead of declaring done
• Resumed session's reply sounds like "already implemented" → a PR is still opened, issue not closed
• Rework pickup completes successfully → a PR is created from the finished work
• Rework pickup hits an error/usage limit again → issue is re-parked as ai-rework, not silently dropped
• Re-remove ai-rework after a second failure → next pickup resumes the latest session, not the stale original
• Check the park comment on a freshly parked issue → wording says resuming the same session, not "queue a fresh attempt"
• Pick up a genuinely new/fresh issue (never parked) → runs the normal fresh pipeline, unaffected by rework changes
• Rework pickup applies to a bug-type issue → resumes correctly, same as a feature-type issue

[ngthluu]

🤖 PR: #52