Use remove_dir_all crate for more robust Windows CI

[trevyn]

std::fs::remove_dir_all has serious issues on Windows: rust-lang/rust#29497

@XAMPPRocky, who wrote and owns the remove_dir_all crate, is on the Rust release team (see https://www.rust-lang.org/governance/teams/release), notes that the crate is used in cargo and rustup, and suggests moving the crate's implementation into std (rust-lang/rust#29497 (comment)), so I think we can trust this dependency. Given that it improves a real issue on Windows, I think that until this is fixed in std, it would be wise to avoid fs::remove_dir_all wherever possible.

[quentinlesceller]

Good idea though not a big fan of having yet another dependency...

[trevyn]

remove_dir_all 0.5.3 (vs current 0.7.0) is already in the Cargo.lock as a dependency of tempfile, would you prefer to use that version?

It's also only added in dev-dependencies since it's only for tests.

Let's see if it actually works...

[trevyn]

Ok, the Windows test passes and it doesn't look like there are any bug fixes between 0.5.3 and 0.7.0: https://github.com/XAMPPRocky/remove_dir_all/commits/master

Shall I use 0.5.3 instead and replace the other test instances of fs::remove_dir_all with remove_dir_all::remove_dir_all, and keep the Cargo.lock changes minimal?

[quentinlesceller]

Probably better to use the 0.7.0 version then.