Unit-test thread cleanup

[devinrsmith]

During testing of #4756, I discovered that io.deephaven.server.util.Scheduler and io.deephaven.util.reference.CleanupReferenceProcessor threads were leaking. CI testing was throwing "can't create more threads" errors due to the addition of a large number of new java client async tests which pushed the leaked thread counts over some JVM limit.

As concrete evidence of the leaks, I ran

./gradlew java-client-session-dagger:test -PforceTest=true --max-workers 1

jps
jstack <pid-for-GradleWorkerMain> | grep os_prio | awk -F " " '{print $1}' | sort | uniq -c | sort -nr

and got

    432 "DeephavenApiServer-Scheduler-Concurrent-4"
    432 "DeephavenApiServer-Scheduler-Concurrent-3"
    432 "DeephavenApiServer-Scheduler-Concurrent-2"
    432 "DeephavenApiServer-Scheduler-Concurrent-1"
     43 "CleanupReferenceProcessor-liveness-drainingThread"
     18 "GC
     10 "G1
      4 "C1
      3 "C2
      3 "/127.0.0.1:42968
      2 "VM
      1 "Test
      1 "Sweeper
      1 "StrDedup"
      1 "Signal
      1 "Service
      1 "Reference
      1 "process
      1 "PeriodicUpdateGraph-updateExecutor-1"
      1 "grpc-timer-0"
      1 "grpc-shared-destroyer-0"
      1 "grpc-default-executor-6"
      1 "grpc-default-executor-5"
      1 "grpc-default-executor-4"
      1 "grpc-default-executor-3"
      1 "grpc-default-executor-2"
      1 "grpc-default-executor-1"
      1 "grpc-default-executor-0"
      1 "Finalizer"
      1 "Common-Cleaner"
      1 "Attach

This PR adds an unit-test-only shutdown procedure for the Scheduler; if we wanted to actually change the Scheduler interface to support this, it would require more work and testing. This also ensures that CleanupReferenceProcessor are promptly cleaned up (instead of eventually).