Test MMAP Timing Model accuracy

First measure RT latency.
Then start with a stable loopback glitch test with N buffer.
Then slowly increase a sleep time inside the callback to see if you can measure glitches with **no** increase in the XRunCount. This works around the problem of AAudio enforcing a minimum BufferSize. This detects if the DSP reads earlier than reported. Adding a negative offset to timestamp should fix it.

We could also test if we can sleep so long that we get XRuns but no detected glitches?

Then setBufferSize to Max to see if there are glitches. If so then search for sleep time that is on the edge of causing glitches. This detects if the DSP reads later than reported. Adding a positive offset to timestamp should fix it.

A similar test could be done for input timing. Use a separate callback thread on a different CPU.