opti audio stream function

[wiiznokes]

Avoid two allocation in the audio stream function

• chunk is a wrapper over 2 slices (a view over the circular buffer), so we can use those instead of allocating a vector

  chunk
                          .into_iter()
                          .collect::<Vec<_>>()
                          .chunks_exact(std::mem::size_of::<F>())

• we collect the sample in a buffer, and then write to the audio buffer. We can avoid this and directly write to the audio buffer

I've not benchmarked my changes tho

[wiiznokes]

edit: i've added benchmarks

     Running benches\audio.rs (target\release\deps\audio-633b2ef21179e806.exe)
Gnuplot not found, using plotters backend
audio_processing        time:   [1.7843 µs 1.8321 µs 1.8847 µs]
                        change: [−43.366% −41.916% −40.626%] (p = 0.00 < 0.05)
                        Performance has improved.

     Running benches\audio.rs (target\release\deps\audio-633b2ef21179e806.exe)
Gnuplot not found, using plotters backend
audio_processing        time:   [3.2307 µs 3.2448 µs 3.2614 µs]
                        change: [+70.207% +73.747% +77.295%] (p = 0.00 < 0.05)
                        Performance has regressed.
Found 25 outliers among 100 measurements (25.00%)
  24 (24.00%) low severe
  1 (1.00%) high mild

above is with the optimization made in this patch, under is master, adapted to the test.
We can see a nice 50% performance boost!

[teamclouday]

Nice thanks for the optimization! Yes it makes sense the extra allocation is slow and unnecessary