chore(args): introduce a more generic solution for restoring double dashes in args

[KuSh]

Summary

• Introduce a shared, generic solution to restore double dashed into args
• Replace custom solutions in cargo and git commands with that solution

Fixes #1669

Test plan

• cargo fmt --all && cargo clippy --all-targets && cargo test
• Manual testing: rtk git diff -- "src/core/args_utils.rs" output inspected

[truffle-dev]

Tested at 4153ebb against the original #1669 repros. All cases I'd tracked work:

• rtk git diff -- <file> (single file after --) — passes
• rtk git diff -- file1 file2 file3 (multiple files after --) — passes
• rtk git diff HEAD -- <file> (ref before --, file after) — passes
• rtk git diff <branch> -- <file-named-same-as-branch> (file shares name with a ref) — passes
• rtk git diff --cached -- <file> — passes
• rtk git diff (no args) — passes
• rtk git diff <bare-ref-with-no-dash> — passes

cargo test --release args_utils is 15/15 green; the unit tests cover the swallowed-by-clap, preserved-by-clap, no----in-original, and interleaved----and-args shapes.

One behavioral note worth flagging in case it's intentional: with the old normalize_diff_args heuristic, rtk git diff main (where main is both a branch and a file on disk) would have had -- injected based on the filesystem check and rendered the file diff. With this PR, the same command falls through to git which then exits with fatal: ambiguous argument 'main'. That looks like the right call to me — restoring what clap consumed is predictable, second-guessing the user with filesystem heuristics is not — but flagging it explicitly so the design choice is on the record.

Happy to close #1679 once this lands. Want me to do it now or wait until merge?

[KuSh]

Thanks a lot for the detailled test and report, very much appreciated!

One behavioral note worth flagging in case it's intentional: with the old normalize_diff_args heuristic, rtk git diff main (where main is both a branch and a file on disk) would have had -- injected based on the filesystem check and rendered the file diff. With this PR, the same command falls through to git which then exits with fatal: ambiguous argument 'main'.

The problem would have happen without rtk and should be handled by the LLM IMHO

Happy to close #1679 once this lands. Want me to do it now or wait until merge?

As we have a release branch, merging doesn't mean available in a release, we will close it once released. Thanks again!