Add double buffering to RawSample

[timchinowsky]

This PR contains part of the changes necessary to implement the processing of live sampled audio as described in #2676. Here audiocore.RawSample is changed so that the sample can present itself either as a full-length single-buffered object or as a double-buffered object comprised of two half-length buffers.

[tannewt]

One question. Thanks for the PR!

[timchinowsky]

Revised RawSample will not loop unless played with loop=True, otherwise behavior is unchanged. Test code exercises this with an audiomixer with 3 voices. Here's how you run the test for all combinations of single/double buffer and loop/not loop:

>>> mix.play(single_buffer=True, loop=True)
(200, 1, 0)
(400, 0, 1)
(100, 1, 1)
>>> mix.play(single_buffer=False, loop=True)
(200, 1, 0)
(400, 0, 1)
(100, 1, 1)
>>> mix.play(single_buffer=True, loop=False)
(200, 1, 0)
(400, 0, 1)
(100, 1, 1)
>>> mix.play(single_buffer=False, loop=False)
(200, 1, 0)
(400, 0, 1)
(100, 1, 1)

Looking at the size and timing of the buffers pulled in by audiomixer, here's what it looks like looped single-buffered transfers:

The pattern repeats every 10000 us, as you'd expect for an 80 sample audiomixer playing at 8 kHz:

With double buffered transfers, you see twice as many transfers that are half as big (and take half the time), which makes sense:

For the examples above, the parameters are chosen such that the audiomixer buffer length is an integer multiple of each of the sample lengths, so the length of the transfers which fill up the buffer are exactly the same each time. For the general case, partial samples will be assembled into the buffer to make up the required number of samples. Here's a plot that shows that behavior. This run used the default buffer size of 128 samples and fills the buffer every 16000 us:

[tannewt]

Great! Thank you!