When offing a voice, the release time is cut to a minimum. Dynamic
updates would reset this to the original value. The commit adds a way to
disable dynamic EG updates and does so when offing the voice.
The engine will also steal the oldest offed voice as a last resort to
hopefully sound the newly played note every time.
Updated tests mostly to call on `renderBlock()` before testing things like number of active voices and such, in case we buffer/reorder events in the library later on.
