Opti audio stream processing

[wiiznokes]

There is a lot of allocation / potential optimization we can do in this hot path.

The benchmark on master is

bench_process time: [123.76 µs 123.78 µs 123.80 µs]

Lets see where we can go

[wiiznokes]

So, in the end, there is no significant performance gain.
On linux, we see a real gain for resampling, when updating rubato to version 2.0 (15.407 µs to 13.536 µs in average).
This is because the new rubato release provide a way to not copy the input and output in a buffer.
The other change are honestly not significant. But i don't think that they complicate the code.

Also, the update of rubato fix some setup, like input = 41000, output = 41000, speexdsp activated.

[teamclouday]

Sorry just saw this. Thanks for the updates!