Add subtree filter support and improve round-trip check

[makai410]

Fixes: #37

To explain what the subtree filter is, here's an example with rustc_public:

org = "rust-lang"
repo = "project-stable-mir"
filter = ":/compiler/rustc_public:exclude[:[::build.rs,::tests/,::Cargo.toml]]:prefix=rustc_public"
subtree-filter = ":/rustc_public:exclude[:[::build.rs,::tests/,::Cargo.toml]]:prefix=rustc_public"

The filter is applied to the rust-lang/rust side, while subtree-filter is applied to the rust-lang/project-stable-mir side. Ideally, both should produce the same SHA, confirming a successful round-trip.

For repos doing a 1:1 mirror, subtree-filter should be empty, in that case this PR will not download the josh-filter tool.

And basically, the round-trip check in this PR looks like:

$ git fetch <josh_url with the `filter`> <branch>
# if the `sym-filter` is set
  # install the josh-filter tool
  $ cargo install josh-cli --git https://github.com/josh-project/josh.git
  $ josh-filter "<subtree-filter>" HEAD
  $ git rev-parse FILTERED_HEAD
  $ git rev-parse FETCH_HEAD
  # then compare FILTERED_HEAD and FETCH_HEAD
# else
  $ git rev-parse HEAD
  $ git rev-parse FETCH_HEAD
  # then compare HEAD and FETCH_HEAD

[Kobzol]

We should probably start using the new Josh CLI in josh-sync, I feel like we are reimplementing a lot of stuff in josh-sync that is already there as a single command :) In any case, if this helps your use-case, it seems fine.

I would maybe rename sym-filter to subtree-filter or something, to make it more explicit about what it does, but it doesn't matter much, sym-filter is fine too.

[Kobzol]

Can you please pin the version/tag/commit SHA? I don't want josh-sync to install unpinned versions of Josh, we had enough trouble with that in the past.

[Kobzol]

Why was a new function needed here?

[makai410]

It's basically a copy-paste of run_command_inner but accepting a PathBuf as the command. But yeah I see that's kind of duplicate, I'm going to fix it.

[Kobzol]

Suggested change

	pub sym_filter: Option<String>,
	#[serde(default, skip_serializing_if = "Option::is_none")]
	pub sym_filter: Option<String>,

Otherwise the config would fail to be parsed if the field is missing, I think.

[makai410]

I would maybe rename sym-filter to subtree-filter or something, to make it more explicit about what it does, but it doesn't matter much, sym-filter is fine too.

Totally, I agree that subtree-filter is way more clear than sym-filter. sym-filter is kind of ambiguous without taking a look at its definition :)

[Kobzol]

Thanks! Let's hope this doesn't break anything :)