Indicate dirty state in short-format submodule diffs

[MsfPablo]

Summary

When diff.submodule = short (the default), a dirty submodule shows:

-Subproject commit ca030fd1a02225a6fc1a834c480276d9c97a8c6f
+Subproject commit 803be42ca46af0fbc65b54a9abfb499389516939-dirty

get_submodule_short_commit matched the optional -dirty suffix but only
ever returned the hash capture group, discarding the flag. So delta
rendered a dirty submodule identically to a clean one — just
<red hash>..<green hash> — which is exactly the ambiguity described in #1838.

This propagates the dirty flag through and appends -dirty to the
printed commit range when the working-tree side is dirty:

ca030fd1a022..803be42ca46a-dirty

Test plan

• cargo test — 438 passed, 0 failed
• Updated test_simple_dirty_submodule_diff to assert the -dirty suffix is rendered
• cargo fmt --check clean
• cargo clippy --all-targets — no new warnings introduced (pre-existing warnings are in unrelated utils/process.rs)

Fixes #1838

[MsfPablo]

Following up on this one. It surfaces the dirty marker in short-format submodule diffs, matching what git itself shows. Glad to add a test case if that'd help review.