Add audio_async_render

[Ichunjo]

This PR adds audio_async_render. It works the same way as clip_aync_render, requesting frames asynchronously using get_frame_async.

I had to modify a part of the code to make it work like renaming RenderContext and finish_frame.
I tried to add a lot of comments to better understand the code since I had to document myself to write the function.

I think adding the URLs I used in the code wouldn't be best idea so here are them:

https://web.archive.org/web/20051028052230/http://media.vcs.de/Downloads/Sony_Wave64.pdf
https://github.com/vapoursynth/vapoursynth/blob/master/src/common/wave.cpp
https://github.com/xiph/flac/blob/27c615706cedd252a206dd77e3910dfa395dcc49/src/flac/encode.c
https://github.com/tpn/winsdk-10/blob/master/Include/10.0.14393.0/shared/mmreg.h#L2107
https://docs.microsoft.com/en-us/windows/win32/api/mmreg/ns-mmreg-waveformatextensible
scipy/scipy#6852
https://gist.github.com/josephernest/3f22c5ed5dabf1815f16efa8fa53d476
https://stackoverflow.com/questions/31739143/when-to-use-wave-extensible-format

Why in lvsfunc?

Well there is already clip_async_render and there is relatively duplicated code. I didn't change anything inside clip_aync_render but changes could be made now that audio is a thing in Vapoursynth.

Why numpy?

A vapoursynth.RawFrame is a Sequence like of memoryviews. We need to stack them to perform computations. I don't know if a solution without numpy exists but it would probably be slower.
Numpy is used quite enough in python environment and some vapoursynth *funcs need it. Plus it's a pretty solid package highly maintained.

Performance

I still need to do proper performance comparisons but it's ~40 % slower than vspipe. Makes sense since it's done in python and not in C.

24:28 minutes 24 bit wave file:

• audio_async_render Duration: 16.13 seconds
• vspipe Duration 11.20 seconds

[Ichunjo]

Mypy's complains is the out of date stubs. They don't include audio stuff yet. I could update them but I think louis used his fork to generate them.

[Ichunjo]

Performance tests:

# audio_async_render

# Wav 16 bits
# Best duration: 9.78
# Mean duration: 13.22

# Wav 24 bits
# Best duration: 15.34
# Mean duration: 19.14

# Wav 32 bits
# Best duration: 17.22
# Mean duration: 22.37

# vspipe

# Wav 16 bits
# Mean duration: 7.86

# Wav 24 bits
# Mean duration: 11.25

# Wav 32 bits
# Wave64 header -> Mean duration: 15.14
# Wave header -> Mean duration: 20.49

Made on a 24:28 minutes wave file

[Ichunjo]

Added in Ichunjo/vardautomation@a334b2c using clip.frames(...) instead. Speed isn’t impacted, thanks to the updates of the function.