diSPIM: update the acq eng to match SCAPE's finish() method, fixes some bugs for both engines along the way

[bls337]

Summary

Ported finish() lifecycle cleanup from AcquisitionEngineScape to AcquisitionEngineDispim for parity between the two acquisition engines. diSPIM's finish() was previously almost a no-op (just polling restart) — the equivalent cleanup already existed, but was inlined at the tail of run() using local variables that weren't reachable from finish(). Promoted that state to instance fields (controller_, savedExposures_, isShutterOpen_, autoShutter_) and moved the cleanup into finish() so it now runs reliably regardless of how the acquisition ends.

New in diSPIM's finish()

• Stop any still-running camera sequence acquisitions
• PLogic/controller cleanup (cleanUpControllerAfterAcquisition, stopSPIMStateMachines)
• Shutter/autoshutter restore
• Check for acquisition exceptions
• Camera trigger-mode reset to INTERNAL + exposure restore
• unregisterForEvents(this) — diSPIM registered for events on every acquisition run but never unregistered (listener leak); now fixed

Bugs found along the way (fixed in both engines)

1. NPE risk in PLogic cleanup: SCAPE's finish() called controller_.cleanUpControllerAfterAcquisition(...) with no null check. In demo mode, controller_ is never instantiated even when "use PLogic" is enabled in settings — so finish() would NPE right there and silently skip everything after it (shutter restore, exception check, camera reset). Fixed by guarding with controller_ != null in both engines.
2. NPE risk in exposure restore: savedExposures_ had no initial value in either engine, only getting populated near the top of run(). If run() threw before reaching that point, finish() would still execute (via the outer finally) and NPE on `save