Allow to use float output in any case if output provider doesn't support 16-bit int

[nift4]

This commit is a continuation of 4f4ad5f and fe4f235 by Toni, decoupling DefaultAudioSink logic from assumptions that AudioTrack is used.

• If audio output provider doesn't support 16-bit int PCM, we should use float PCM in any case, even if input isn't high-resolution PCM
• Unlike when this check was written, ToFloatPcmAudioProcessor now supports every linear PCM format, similar to ToIntPcmAudioProcessor, it is no longer limited to high-resolution only
• While AudioTrack (and by extension AudioTrackAudioOutputProvider) always supports 16-bit int (hence, this change is a no-op for them), it may not apply to other ways of outputting PCM
• Additionally, the same applies to the provider possibly not supporting float PCM output. Currently, we have the enableFloatOutput option, but this is scheduled to be replaced by a more generic one. While we're here, we can address it as well by adding a check.

[tonihei]

An audio output not supporting 16-bit PCM sounds very unusual. I appreciate that the abstraction of the AudioTrack-based output makes this theoretically possible, but I wonder if you have a concrete use case in mind?

(PR looks good otherwise, just making sure we really need this bit of logic)

[nift4]

I was thinking of a custom PCM mixer for my WIP libusb output and that should use the highest determinator format. A PCM mixer can only mix what has the same format, and I intend to do with float32 for being said highest determinator. So I need to convert all audio to float32.