WIP/RFC: Use Artifacts system

[sjkelly]

This switches GR to use the Artifacts system in its entirety, leading to substantially improved load times.

Before:

julia> @timev using GR
  0.986996 seconds (844.83 k allocations: 49.678 MiB, 1.45% gc time, 88.20% compilation time)
elapsed time (ns): 986996340
gc time (ns):      14332472
bytes allocated:   52091322
pool allocs:       844390
non-pool GC allocs:434
malloc() calls:    6
GC pauses:         1

After:

julia> @timev using GR
  0.243714 seconds (406.49 k allocations: 24.254 MiB, 7.21% compilation time)
elapsed time (ns): 243714086
bytes allocated:   25431816
pool allocs:       406055
non-pool GC allocs:360
malloc() calls:    67
realloc() calls:   7

The substance of the change is quite small, though I suspect some documentation and notes will be required to update workflows (e.g. how to use Overrides.toml rather than environmental variables).

I am throwing this up as an RFC, as this mostly will remove the ability to use environmental variables to select the provider, and rather use Artifacts. As noted, this provides a faster TTFP, but I am unsure if there are concrete reasons (aside from legacy compatability) that environmental variable support should remain.

[mkitti]

By removing access to alternate binaries, we removed Wayland support.

sciapp/gr#141