Allow overriding the dist server

[roblabla]

The goal of this PR is to make it possible to use a mirror of rustup, such as one created using Panamax, with fenix.

To do so, it fixes three issues:

• When setting a custom dist server, if the manifest contains links to the default dist server, rustup will automatically replace it with an URL pointing to the custom dist server. This allows making very naive mirrors of the rustup dist server without having to modify the channel manifests.
  Currently, fenix allows setting a custom dist server through toolchainOf, but it does not replicate this behavior of overriding the URL in the manifests. This PR adds this behavior.

• Furthermore, it adds a root argument to fromToolchainFile, allowing the simultaneous use of a toolchain file and a custom dist server.

• Finally, it replaces the archive it gets from the gz ones to xz ones. Rust started providing xz archives in may 2017, and rustup switched to grabbing the xz archives not long after. Panamax, the most popular tool to mirror rustup, defaults to only grabbing the xz archives.

[roblabla]

Bump?

[roblabla]

Bump? Is this project dead?

[eureka-cpu]

Nice one @roblabla. This looks great, and doesn't change the default behavior. Would you mind updating the README with this use case as well?

[eureka-cpu]

Also please target staging.

[roblabla]

I added a quick documentation line for the new root parameter in the README. I can provide a full-blown usage example for the use-case if necessary - adding it under the # Examples heading?

[eureka-cpu]

I added a quick documentation line for the new root parameter in the README. I can provide a full-blown usage example for the use-case if necessary - adding it under the # Examples heading?

Thanks, that would be great 🙂

[eureka-cpu]

@roblabla is this ready? if not just request my review again when you are 🙂

[roblabla]

CI is failing, need to look into it. I think it's broken because the hardcoded toolchain URLs in data only lists the url/hash, and not the xz_url/xz_hash.

I'll need to either

1. make a script that swaps it from the former to the latter (and update whatever script is responsible for updating it to grab the xz)
2. Make the code fallback to using the gz if xz is absent.

[eureka-cpu]

Nice one. Probably we will want to test it in CI with something small, so we aren't trying to build an entirely separate toolchain which isn't cached with the substituter. I think this way we can also be sure if it's actually using the xz_url.