Tracking issue for audio engine rewrite

[yara-blue]

This tracks landing the changes from https://github.com/RustAudio/rodio-experiments into Rodio. This involves integrating the new rubato resampler and specializing it where possible.

Background:

For a long time we've had mysterious bugs round spans. Spans allow a source to change parameters (sample-rate and channel count) during playback. We've concluded that spans where a mistake as correctly handling audio that at any point is:

• really hard to do correctly.
• impossible to optimize well leading to bad performance.

Most audio should be converted to the playback parameters at the very start of the audio pipeline. To make this easier we are adding two new source types. Both do not allow parameters to change, one fixes this at runtime the other needs the parameters known at compile time.

Approach

This is a large overhaul at around 7700 lines of code (ignoring docs & comments). To keep things reviewable we'll do this in reasonably sized PRs of around 500 lines. Help is greatly appreciated!

Phase one

Implementation in: #903 and #904

• land const source
• remove experimental from fixed source
• land try-into adaptors/convertors between sources (others in phase two)
• note overhaul in Readme
• land most basic generators

Phase two

Implementation in: #905 and #910

• channel conversions
• resampler and integrate with rubato

Phase three:

Implementation in: #911

• land one effect and pure_effect! macro
• land all other effect
• deprecation warning on effects in DynamicSource that are not sound

Phase four

• add new combinators [a, b, c].mixed() etc.
• add all generators (noise etc, function gen. etc)

Phase five

• rename current source to dynamic source
• add Source = DynamicSource type alias with deprecation warning
• Remove placeholders

Phase six

• port all examples
• re-add controls to automatic gain control
• rewrite Rodio main docs
• update porting guide (needs to be really nice this time :3 )
• prepare and make announcements

Open Questions

• Do we deprecate Source in favor of DynamicSource?
• Do cons sources and the ConstSource trait get default generics?

[yara-blue]

One point for discussion:

Should the ConstSource have a default generic? as in should it be:

trait ConstSource<SR=44100, CH=2> { ... }

Such that it can be used as:

fn takes_a_const(source: impl ConstSource)

Or

fn takes_a_mono_const(source: impl ConstSource<CH=1>)

If we do we should do the same for the generators (for now only silence but we will land others soon see #901)

For now I've gone with no default generics. While removing them would be a breaking change adding is not*, so we can safely postpone the decision.

*: this gets complicated but let's just say it is not breaking...

[yara-blue]

Since this is a rather big effort I'd love to spread out the review load a little. @PetrGlad do you have time by change to review some of this? Each PR is around 500 lines when compared to the previous (so 1a compared to 1b etc). I'd also love your take on this @iluvcapra.

The larger effort is to add two new source traits that both forbid sample rate and channel count from changing. This enables a bunch of optimizations (or rather saves us from doing a bunch of checks) and makes Rodio simpler and more correct. In time we will be moving over functionality from the current Source trait to these two new traits: (FixedSource and ConstSource).