midi.config.add_synth now subsumes add_synth_object.

[dpwe]

Originally we had midi.config.add_synth() which took a patch number and voices count.
Then we created the Synth() object which handled the patch number. But add_synth called it.
Then we added add_synth_object so you could construct your own Synth object and add it to MIDI.
But users were confused about how synths were created since midi.config.add_synth() was doing it for them.

This change alters midi.config.add_synth() to accept a Synth object, and to prefer that pattern. This makes it obvious to the user that Synth objects are the core concept.

The problem here, I realize, is that in order to manage the limited voices resources, it's best to release any existing synth before creating the new Synth object. However, if the Synth is passed as an instantiated object it will already have grabbed its voices. Maybe we can fix with some kind of deferred initialization in the Synth object.

[dpwe]

I added deferred initialization to the Synth() object, so the midi.config.add_synth(synth=Synth()) will cleanly reuse the voices from any synth it replaces.

[bwhitman]

This LGTM and does make more sense, i think