Nuke `CARGO_*` env variables for nested cargo

[khyperia]

I hit this in #291 and did a quick fix with my Try workaround on CI... commit and was intending on doing the proper fix if that passed, but mergify got happy and did its thing. So, filing this issue so we remember to do the proper fix.

Not sure if we should nuke all CARGO_* env variables, we should look through the list and figure out which ones to keep, but generally yeah.

More explanation on how this causes things to go wrong can be found in that link to the glam PR in #291

[XAMPPRocky]

I have a fix for this on the vuild-system branch, by just removing all env variables and putting PATH back in.

[khyperia]

clearing all environment variables sounds incorrect, there's potentially really important variables configuring things, right?

[XAMPPRocky]

clearing all environment variables sounds incorrect, there's potentially really important variables configuring things, right?

Well I should be clear, all except the ones we set. So we're still passing arguments and RUSTFLAGS to configure cargo/rustc. It is working for me on all the runners, and fixed the clippy issue for me.

[eddyb]

I don't think it's a good idea to remove more than CARGO_FEATURES_* and CARGO_CFG_* (which we know the nested Cargo will set whichever are appropriate - also, AFAICT there are no other Cargo env vars that have dynamic names like that).

It might work in isolation, but when taking into account all OSes and Linux distros, and whatever build scripts might be running for dependencies of shaders, etc. I don't see it always working, and it's safer to only remove the problematic sets.