fix(tracker-intake): keep issue claimed while a terminated worker's PR is in flight

[illegalcall]

Problem

#2777 fixed #2746 by excluding terminated sessions from intake's seen set, so a genuinely dead worker no longer blocks re-spawn. But that opened a new window (#2921):

• Intake dedups only on live sessions (seenIssueIDs) and has no PR awareness — its Store interface never exposed PR facts.
• Manager.Spawn has no issue-level dedup.
• A Fixes #N PR closes issue #N only on merge, not on open.

So when a worker opens a PR and is then killed or crashes while its issue is still open (PR unmerged), the terminated session drops out of seen, and the next 1-minute tick spawns a duplicate worker — and a second PR — for the same issue. The simplest trigger needs no crash: an operator killing a worker that already opened a PR.

Fix

Intake now also treats an issue as claimed when a terminated session for it left an open or merged PR, using the existing per-session PR store (ListPRFactsForSession). A terminated session keeps its pr rows because kill/terminate only sets is_terminated=1 — it does not delete the sessions row (that persistence is the same basis as #2746), and pr.session_id cascades only on row delete.

seen := seenIssueIDs(sessions)               // live sessions claim outright
o.claimIssuesWithInFlightPR(ctx, sessions, seen) // + terminated sessions with an open/merged PR

Behavior:

• Open or merged PR on the terminated session → issue stays claimed, no duplicate spawn.
• No PR, or only a closed-unmerged (rejected) PR → issue is freed for re-spawn, preserving bug(tracker-intake): terminated sessions block re-spawn of their issue (intake never re-assigns killed/crashed issues) #2746 and allowing a fresh retry.
• PR-lookup error → fails safe toward claimed, so a transient read cannot cause a duplicate PR.
• Only terminated sessions whose issue is not already claimed by a live session are queried, so a healthy project (one live worker per issue) issues no PR lookups.

No schema/query/codegen change — reuses ListPRFactsForSession and adds one method to intake's local Store interface (satisfied by the production *sqlite.Store).

Tests

Added to the trackerintake package:

• TestPollDoesNotRespawnIssueWithOpenPRFromTerminatedSession — the tracker-intake: duplicate worker/PR spawned for an open issue whose worker was terminated after opening its PR (regression window from #2777) #2921 case.
• TestPollDoesNotRespawnIssueWithMergedPRFromTerminatedSession — merged work isn't redone.
• TestPollRespawnsIssueWhenTerminatedSessionPRClosedUnmerged — rejected attempt retried.
• TestPollTreatsIssueClaimedWhenPRLookupFails — fail-safe.
• TestPollLiveSessionClaimsWithoutPRLookup — live session claims with no PR query.

All existing tests (including #2777's) pass unchanged; full package + go build ./... + go vet clean.

Scope

Only touches trackerintake's dedup logic and its tests. Does not change the SQL layer, session lifecycle, or Spawn. Distinct from #2745 (reaper cannot terminate stuck sessions).

Fixes #2921

🤖 Generated with Claude Code

[somewherelostt]

Thanks for contributing to Agent Orchestrator.

This PR is being picked up by the current external contributor on-call pair:

• @illegalcall
• @Pulkit7070

If someone is already working on this, please continue as usual.
The on-call pair is added for visibility, tracking, and support, not to take over the work.
If you need help with review, direction, reproduction, or next steps, please tag @illegalcall and @Pulkit7070 here.

For faster context or live questions, you can also join the AO Discord.

Join the session here:
https://discord.gg/H6ZDcUXmq

Come by if you want to see what is being built, ask questions, or just hang around with the community.

[AllBeingsFuture]

Superseded by #3026 — same intake fix for #2921, plus cherry-pick of #3014 migration renumber so CI can pass (main contamination: goose duplicate version 25). Could not push to original head branch (no write access).