Exit cleanly when startup fails instead of leaving a zombie JVM (fixes #498, fixes #499)

Problem

MirthLauncher loads com.mirth.connect.server.Mirth as a Thread and starts it; it never
calls Mirth.main(), so main()'s System.exit(0) never runs in the real launcher path.
When run() aborts a failed startup it simply returns — the Mirth thread dies, but the
launcher's non-daemon threads keep the JVM alive, so the process stays up with no listening
ports. A service manager sees it as running; nothing restarts or alerts (#498).

Two confirmed triggers:

• A port already in use (e.g. a second instance, or another service on 8443).
• A failed database migration (MigrationException).

On the migration path there is a second problem (#499): the shutdown hook, registered before
startup(), runs the full Mirth.shutdown(), which reads the database while stopping the
engine. Startup already called stopDatabase(), so on Derby the hook waits for a pooled
connection that never comes and hangs — the process must be SIGKILLed.

Fix

• System.exit(EXIT_STARTUP_FAILURE) on both abort paths (port-in-use and MigrationException)
  so the JVM actually terminates with a non-zero status a service manager can see.
• Track startupComplete; the ShutdownHook runs the full shutdown() only when startup
  completed. Exiting after a failed migration therefore skips the engine/database shutdown that
  would hang on Derby.

Normal operation is unchanged: a server that finished starting still shuts down fully via the
hook and via the SHUTDOWN_SERVER command.

Verification

Built from this branch and run live (Open Integration Engine 4.7.0, embedded Derby, in Docker):

Port already in use — start a second instance while 8443 is taken:

• Before: the second JVM stayed alive (orphan process, no listeners); the wrapper had to be timed out.
• After: exits code 1 in ~2s, no orphan process.

Failed migration — set SCHEMA_INFO.VERSION to an unknown value, then start:

• After: logs MigrationException: Database schema version "99.0.0" is not recognized ... and the
  process exits code 1 in ~5s (no Derby shutdown hang, no SIGKILL needed).

(The pre-fix port-in-use zombie was reproduced directly; the migration-path Derby hang matches the
behavior documented in #499 and shares the same root cause.)