IpmiClient timeout: cleanup runs alongside the still-running worker, and Close Session is never confirmed #132

Where: src/main/java/org/metricshub/ipmi/client/IpmiClient.java (try-with-resources around Utils.execute), src/main/java/org/metricshub/ipmi/client/Utils.java (execute), src/main/java/org/metricshub/ipmi/client/runner/AbstractIpmiRunner.java (close).

Context: Codex raised this on #124: Utils.execute() only bounds the runner's call(); the try-with-resources in each IpmiClient method then calls AbstractIpmiRunner.close() on the calling thread, after the deadline, so the cleanup is not covered by IpmiClientConfiguration.timeout.

What the code does today:

• Utils.execute() cancels the future (cancel(true)) and calls shutdownNow() without waiting for the worker. Because Connection.waitForResponse() swallows InterruptedException (Timeouts and cancellation: waitForResponse counts sleeps instead of wall-clock, swallows InterruptedException, and non-daemon threads keep the JVM alive #79), the worker usually keeps running.
• close() then runs on the calling thread while the worker is still using the connector: closeSession() sends Close Session without waiting for its reply (the SessionValid state switches to Authcap and sends the request), and tearDown() closes the UDP socket and the timers under the running worker, which then fails and logs errors.
• In the current code nothing in close() waits for a reply, so the caller's TimeoutException is not delayed in practice. But this is not guaranteed by design or by any test, and:
  • Close Session is never confirmed: if it is lost, the BMC session stays allocated until it expires;
  • the worker thread can outlive the call and keep the JVM alive (Timeouts and cancellation: waitForResponse counts sleeps instead of wall-clock, swallows InterruptedException, and non-daemon threads keep the JVM alive #79);
  • close() throws NullPointerException when startSession() failed before the connector was created (Timeouts and cancellation: waitForResponse counts sleeps instead of wall-clock, swallows InterruptedException, and non-daemon threads keep the JVM alive #79).

Suggested fix:

• make the cleanup part of the deadline explicitly: stop the worker (interruptible waits, Timeouts and cancellation: waitForResponse counts sleeps instead of wall-clock, swallows InterruptedException, and non-daemon threads keep the JVM alive #79), wait for it a bounded grace period, then close;
• confirm Close Session with a short timeout taken from the remaining budget (best effort, never beyond the deadline);
• make close() null-safe and idempotent;
• add a test against a fake BMC (a local UDP socket that answers the handshake and then drops replies) asserting that every IpmiClient method returns within timeout plus a small margin, and that no library thread survives.

No documentation change is needed until this is fixed.

Related: #77 (per-message timeout), #78 (sync retry), #79 (cancellation, non-daemon threads).