Use explicitly defaulted destructors for base cases when possible,
eliminate aggregate initialization constructors where possible, and push
some more stuff to NSDMIs.
Simply zero the memory for audio which hasn't been decoded yet, modify
the audio renderer to avoid caching blocks which aren't ready yet, and
add a progress indiciator to the audio display scrollbar.
Register functions which create each type of provider rather than the
provider types themselves so that the concrete types don't need to be
publicly exposed, and use a static list of providers rather than
registering them at runtime.
