CI: isolated UT Coverage package exceeds 10-minute deadline during cluster shutdown #29368

Summary

The pkg/tests/issues/isolated package exhausts its 10-minute go test package deadline in UT Coverage across several unrelated open PRs. The package budget is shared by multiple embedded-cluster tests, so the final running test name is not by itself the cause.

Evidence

• #29365 UT Coverage job timed out after about 601 seconds; the uploaded raw test report shows TestAutoIDCacheRestartAndDisabledNode taking about 302 seconds and TestIssue28234ExceptAllCluster about 157 seconds. TestSpeedupAbortAllTxn had run only 27 seconds when the package deadline fired.
• At the timeout, TestSpeedupAbortAllTxn was cleaning up an embedded cluster. The main shutdown path was embed.cluster.Close -> cnservice.closeService -> Stopper.Stop; a heartbeat task was in createTaskService -> registerExecutorsLocked -> mysqlTaskStorage.PingContext -> net.Read.
• Similar package-level coverage timeouts occurred on fix: cancel stale logtail subscription pulls on disconnect #29361, fix: preserve TRIM and POSITION syntax in CTAS #29359, and fix: support bounded XPath aggregates and MySQL predicate context #29362. Coverage merge then reports that a required producer did not publish a complete profile; that is a downstream failure.
• Related [Bug]: CI UT timeout failures in transaction abort and backup protection tests #25997 records short RPC/checkpoint deadline failures under CI contention. This issue tracks a distinct package-level 10-minute coverage deadline and shutdown wait signature.

Investigation and acceptance

1. Reproduce or compare against clean main using the same UT Coverage mode and package parallelism; collect per-test timings and the full shutdown goroutine graph.
2. Determine whether the heartbeat MySQL I/O can outlive CN shutdown and, if so, make the owner cancel/close it so Stopper.Stop terminates. Do not infer that network read is the root cause from one stack alone.
3. Reduce or isolate the expensive embedded-cluster fixture work so the package finishes with margin under its configured deadline, while preserving each distinct test contract. Avoid simply raising the timeout without resolving the shutdown path and budget.
4. Verify the affected lifecycle case and the owning package in coverage mode, then a complete UT Coverage producer run.