Support sending voice messages about hikari

[hypergonial]

I'm pretty sure the waveform= kwarg should be added to attachments, unless I'm mistaken.

Also, based on this, should waveform= be added to all hikari resource base types? Or should a new basetype be created?

[PythonTryHard]

I vouch for creating a new base type, something like VoiceMessage. Adding waveform= to all hikari resource bases sounds...iffy. I can't describe why, but intuition says it's iffy.

For implementation, I found librosa that does the downsampling in Python rather than having to offload to ffmpeg. The only pain is that librosa only does sample rate, not bit depth, though I assume with some nparray shenanigan the bit depth should be easy enough. It's just a graphical representation, not some audio nerd stuffs.

Although, considering librosa's dependencies, if we were to go with it as a "battery included", I'd rather we put it as another extra like hikari[voicemsg]. How many times do you see a Discord bot library ship with number-crunching dependencies?