PARTIALS timebase dependent on BLOCK_SIZE

[Dan Ellis]

When I changed the BLOCK_SIZE from 256 to 64 samples everything was more or less the same (excepting the 192 sample advance of the envelope due to the reduced impact of the one-frame envelope delay) except for TestPartials, which was completely different. Inspecting the waveforms and spectrograms, there's something weird going on - the shorter BLOCK leads to faster rendering, but it's not a simple linear speedup.
Reference (BLOCK = 256):

New (BLOCK = 64):

[Dan Ellis]

This resulted from partials.c allowing only one new partial to start each frame. By starting as many partials as needed each frame, the gross timebase differences between BLOCK=256 and BLOCK=64 is gone. Fixed in f582fe0.