fixed git submodule bug when using relative paths

[Choudhry18]

Summary

This fixes installations using git which involved submodules declared through relative paths, fixes #9822 .

Test Plan

I did testing using a private gitlab server that I was running locally, to recreate the issue. I have also added tests that involve installing from a repository that I created on my GitHub. Adding that repository and it's accompanying submodule to uv-test (I see there is already a repository - uv_submodule_pypackage that tests submodule installation but that doesn't use relative path) should work.

[charliermarsh]

This looks reasonable and I greatly appreciate the test. My only concern is that Cargo does it in the order we have here 🤔 Can you briefly check if there are any such issues reported in Cargo?

[charliermarsh]

They do have one test here -- I wonder why it succeeds? https://github.com/rust-lang/cargo/blob/bc577dc6c4d65b6d023d08cae20f682a12d85c0c/tests/testsuite/git.rs#L926

[Choudhry18]

They do have one test here -- I wonder why it succeeds? https://github.com/rust-lang/cargo/blob/bc577dc6c4d65b6d023d08cae20f682a12d85c0c/tests/testsuite/git.rs#L926

I looked into Cargo’s implementation and found that their tests do not fail due to differences in how submodules are updated.

Currently, uv relies on the Git CLI (git submodule update --recursive) to update submodules, whereas Cargo passes the remote URL to a submodule function that fetches it using either git_cli, gitoxide, or libgit2. This means that when git reset breaks the link to the remote origin, Cargo’s implementation remains unaffected.

One key reason Cargo seems to run reset first is their guard mechanism, which helps mitigate interruptions while updating submodules and uv's guard mechanism is not affected by this change.

I did think of scenarios where updating submodules before a reset could cause issues—for example, in a local repository with uncommitted changes affecting submodules. To address this, I implemented the fix so that reset is performed before updating submodules.

[Choudhry18]

Apologies for the ping, @charliermarsh . I just wanted to check in to see if you had any questions about the implementation or if there's anything on my end that might be holding up the review of this PR