fix(service): coalesce duplicate EnqueueTaskForIssue on the assignee path (#5914)

[kakiuwang-ui]

Follow-up to #5958, opened at the maintainer's suggestion on #5938:

One narrow difference remains around a concurrent duplicate EnqueueTaskForIssue call. If you would like to pursue that separately, please open a small PR rebased on the latest main, focused on that path and its regression test.

Problem

#5958 mapped the duplicate-pending-task race to ErrDuplicatePendingTask on the
mention path (enqueueMentionTaskWithCommentPlan), but the issue-assignee
path was left untouched. enqueueIssueTaskWithCommentPlan is the second write
point of the idx_one_pending_task_per_issue_agent unique index, so a concurrent
duplicate there still logged at slog.Error and returned a generic wrapped
create-task error — the same HTTP 500 with the leaked constraint name that #5914
set out to remove.

Changes

server/internal/service/task.go

• enqueueIssueTaskWithCommentPlan: reuse the existing isDuplicatePendingTaskErr
  helper and return the existing ErrDuplicatePendingTask sentinel (slog.Debug,
  not Error) on the race — identical treatment to the mention path. No new
  helper or sentinel is introduced.

server/internal/handler/comment.go

• enqueueSingleCommentTrigger (issue-assignee case): the plain slog.Warn was
  left in place by fix(handler,service): map duplicate-key violations to 409/coalesced instead of 500 (MUL-5285) #5958 precisely because this path never returned the sentinel
  yet. Now that it does, switch to the shared logCommentEnqueueFailure helper so
  the benign race is downgraded to debug and resolveCommentTriggerEnqueue
  coalesces — matching the sibling squad-leader and mention cases directly above.

server/internal/service/duplicate_pending_task_test.go

• Add TestEnqueueTaskForIssueCoalescesDuplicatePendingTask next to the existing
  mention-path regression: a second EnqueueTaskForIssue for the same
  (issue, agent) returns ErrDuplicatePendingTask, does not leak the constraint
  name / SQLSTATE, and leaves exactly one pending task.

Other EnqueueTaskForIssue callers (autopilot.go, issue.go,
onboarding_shim.go) are intentionally out of scope: they already logged/wrapped
the pre-existing generic error, so their severity is unchanged; the sentinel is
now available via errors.Is should they want to special-case it later.

Test plan

• go test ./internal/service/ ./internal/handler/ against live Postgres
  (migrated to latest, 251) — all pass, including the new regression test.
• go vet ./internal/handler/ ./internal/service/ clean; changed files
  gofmt-clean.

Refs #5914, #5958, #5938

[vercel]

@kakiuwang-ui is attempting to deploy a commit to the IndexLabs Team on Vercel.

A member of the Team first needs to authorize it.