fix: semaphore release submitted to a dead executor

[triceo]

I ran the original test 100_000 times, it failed zero times.
Claude's summary follows:

---

ConsumerSupport uses 3 semaphores to gate consumeFinalBestSolution(). Each is acquired before a task runs, and released after the task completes via
whenCompleteAsync(release, consumerExecutor).

whenCompleteAsync(action, executor) does not run action inline. It submits action as a new task to executor when the prior future completes. Two-step:

1. Prior task finishes → future completes
2. Java calls executor.execute(releaseTask)

Between steps 1 and 2, executor can be shut down. shutdownNow() rejects new submissions and cancels queued ones. RejectedExecutionException → release task
never runs → semaphore stays at 0.

Who shuts the executor down?

ConsumerSupport.close() — called from DefaultSolverJob.close() when a solver job is cleaned up. Its finally block calls shutdownConsumerExecutor()
unconditionally, even if acquireAll() threw InterruptedException. If the close-thread is interrupted mid-acquire(), the executor dies with semaphores still
at 0 and their release tasks still queued.

Race that produces the hang:

1. Consumer thread finishes start-job task → future F completes
2. Concurrently: close-thread interrupted in acquireAll() → finally → shutdownNow()
3. whenCompleteAsync tries executor.execute(startSolverJobConsumption.release()) → rejected
4. Solver thread later calls consumeFinalBestSolution() → acquireAll() → startSolverJobConsumption.acquire() → blocks forever

Why whenComplete fixes it:

whenComplete(action) (no executor) runs action synchronously on the thread that completed the future — here, the consumer executor's own thread, immediately
after the task finishes, before returning to the executor's work loop. No second submission, no queue, nothing to reject. The release is atomic with the
task's completion from the executor's perspective.

Executor shutdown can only happen after acquireAll() succeeds in consumeFinalBestSolution, which is after all releases have run. The release is now
unreachable by any shutdown path.

[Copilot]

Pull request overview

This PR prevents solver event-consumption semaphores from getting stuck when the consumer executor is shut down by ensuring semaphore releases run inline on the completing thread (instead of being resubmitted to an executor that may already be terminated).

Changes:

• Replace whenCompleteAsync(..., consumerExecutor) with whenComplete(...) for intermediate best-solution consumption to avoid rejected-release tasks.
• Apply the same whenComplete(...) change for first-initialized solution consumption semaphore release.
• Apply the same whenComplete(...) change for start-solver-job consumption semaphore release.

[Copilot]

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated 1 comment.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
100.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[zepfred]

Were you able to reproduce the issue before the fix?

[triceo]

Were you able to reproduce the issue before the fix?

The issue was failing our CI builds couple times a day.