fixes a bug on OpenSL (classic confusion audio sample vs. audio frame)
fixes a crash about logging in OpenSL and AAudio backend (this isn't very good but works well for what we need)
fixes an issue in which cubeb errored out when attempting to read from an input device that was disconnected: this happened when opening a Bluetooth input that was changing protocol or something like that. This is largely the same logic as other backends: device disconnected -> reinit
make it so latency_frames is the buffer size that will be passed to the callback, allowing aligning buffer size in the entire processing chain, yielding improved performance
increase the buffer size to be 3 times the burst size (triple buffering), this seems to behave appropriately, but it's likely that I'll do further testing / tuning. This improves robustness significantly.
This series does the following:
latency_frames
is the buffer size that will be passed to the callback, allowing aligning buffer size in the entire processing chain, yielding improved performance