[CloudCorpRecords]

What changed

Fixes #1828 — the stop-outcome defect.

When several tasks join one run and all are stopped in one batch of concurrent stop calls, a joined task whose stop reached the store after the run's stop was reported as "has already finished; there is nothing to stop", even though its worker was mid-work and was in fact ended by the run's stop (plandb.Store.StopRoot cancelling every open task in the run). The answer depended on the order the concurrent stops reached the store.

• stopJoinedRow (internal/session/stoprun.go): a terminal joined task is now told apart by its ending. Cancelled under a person's stop reason (planStopReason), with the run's root cancelled under the same reason, it answers task N (title) was already stopped with task M (run title), whose run it shared. Cancelled under the stop reason while its run is still open, it answers was already stopped. "Already finished" is now said only for a task that landed or failed on its own.
• endedBeltRow: a settled run row that reads stopped (TaskNotice.Stopped) now answers "was already stopped" instead of "already finished", so a batch stop that arrives after the run has settled agrees with the row it is looking at.

Scope: this PR fixes the stop reporting in the issue (the first defect and the acceptance sentences). The joined card where: folder and the joined receipt's copy wording named in the same issue are not changed here.

How it was checked

New regression tests in internal/session/stoprun_joined_batch_test.go (deterministic: the run's stop is written before the joined stop is issued, and separately all stops are issued concurrently — no timing luck):

• Red on dev (f4db1f4): go test ./internal/session/ -run 'TestStopJoinedRow|TestStoppingRunThenJoined|TestStoppingBatch' -count=1 -v -> 3 failed with the exact reporter sentence task 72 (task D) has already finished; there is nothing to stop; the landed-task guard test passed.
• Green with the fix: same command -> 4 passed.
• Race: same tests with -race -count=3 -> 12/12 passed, no race reports.
• Existing stop family: all 8 tests in stoprun_test.go passed; cancel/stop-boundary tests passed (11).
• Pre-existing failures on this Windows machine, proven identical on a pristine dev worktree at the same commit and unrelated to this change (no plandb code is touched): the plandb StopRoot/Cancel tests fail in TempDir cleanup with plan.json: The process cannot access the file because it is being used by another process (3/3 on both trees), and 5 job/stopwork tests fail with belt has no "bash" tool on both trees.
• gofmt -l on both changed files: clean.
• Full make pr-ready / make test-quick were not run; the focused suites above are what was run, per the repo's focused-test guidance.

Checklist

• A change entry — docs/changes/unreleased/2367-joined-stop-reads-stopped.md, created with make changelog-new PR=2367 KIND=fixed SLUG=joined-stop-reads-stopped in the follow-up commit on this branch (the number only exists once the PR does); changelog-check passes (105 entries, all well formed).
• The manual knows about it — no slash command, key, tool, default, limit or refusal moved; the corrected stop sentence is not quoted in internal/manual/chat/ (searched), and the manual's account of stopping tasks already describes this behaviour.
• Only my own paths are staged — two files, pushed via the Git Data API (no git add -A).

Disclosure: this change was prepared with AI assistance.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[santoshkumarradha]

Hey, thanks for the PR. Could you sign the CLA when you get a chance so we can review it?