Use thiserror and refactor errors

[kvark]

Makes error variants a bit more sane and structured. Also takes advantage of thiserror, since that's what we settled on in wgpu land. PR checklist:

• [x] make succeeds (on *nix)
• [ ] make reftests succeeds
• [ ] tested examples with the following backends:

New code (HAL only):

Finished dev [unoptimized + debuginfo] target(s) in 17.65s

Old code (HAL only):

Finished dev [unoptimized + debuginfo] target(s) in 7.73s

[kvark]

New timing - cargo-timing-20201202T054953Z.html.zip Old timing - cargo-timing-20201202T055308Z.html.zip

It's clear that gfx itself doesn't take more time to build. In fact, even less. But in total the time now has to include the thiserror dependencies, including syn, quote, and proc-macro.

So it's a question of whether we expect the clients to already depend on thiserror (in which case it's free), or not. The benefit is clear - our code becomes more manageable (+338 −865).

[kvark]

I believe the most important (current and future) consumers of gfx-hal are:

• wgpu - already uses thiserror
• amethyst - uses thiserror transitively through cpal, also depends on a lot of other crates that drag syn and friends in
• webrender - also not a concern for it

So I think we should bite the bullet and go for it.

[grovesNL]

bors r+

[bors[bot]]

Build succeeded:

• Android Stable
• Windows Stable
• Ubuntu Stable
• iOS Stable
• Dependency Check
• MacOS Stable