Does not cover the case cargo +nightly b which is due to a missing mtime in the proxy cache.
Workspaces with different toolchains (i.e. with https://github.com/paritytech/substrate ) don't work yet either, possible due to the same root cause.

Is the restriction to rsut 1.36 necessary? That's why the CI fails since Option<Option<_>> does not yet have the flatten.

use a lookup for the compiler_path to a Box<dyn CompilerProxy> which is used for the exact compiler lookup depending on the cwd current working dir.
RustupProxy runs rustup which rustc and creates a Rust / Box<dyn Compiler> instance.
This information is then stored with modifaction time stored as if the Proxy never existed.

Current ugliness:

• compiler type needs to be known so the correct proxy can be picked up - this is all mingled in detect_compiler right now, which returns a Rust and RustupProxy (or more precisely: (Box<dyn Compiler>, Option<Box<dyn CompilerProxy>>)

This was more work than anticipated and the futures pre-stabilization caused quite a few not anticipated changes.
I'd like some feedback on the allover idea, and maybe on how to reduce the clutter a bit.

Verified this at least solves #663 .

@froydnj if you could give some feedback so I can move this forward, that'd be great

@froydnj I addressed all concerns and verified it works using spearow/juice as a demo project - there are still 2 open comments :)

@froydnj can we clarify the last point, (either implement one of both options or merge this as is) unless there are any remaining issues? that'd be much appreciated - thanks!

@froydnj @luser I implemented a hybrid solution of what both you suggested, after all they are not really exclusive and work very well hand in hand. Given the fact that it is only run once per task on the client, the amount of additional work required should be negligible since it's cached for subsequent compiles.

b4aabf0 was added to make +beta happy, also part of #683

Rebased, removed obsolete changes and resolved all review comments.