fix: split dense hunks when staging non-contiguous replacements

[andresespinosapc]

Summary

When the user selects multiple non-contiguous delete+add pairs from a single dense replacement hunk (e.g. file:-2,2,-4,4 against a hunk of -b,-c,-d,+B,+C,+D), the previous implementation collapsed the kept deletions and additions into a single output hunk. git apply rejected the resulting patch with patch does not apply because the deleted lines are not consecutive in HEAD.

Repro

$ printf 'a\nb\nc\nd\ne\n' > dense.txt && git add dense.txt && git commit -m init
$ printf 'a\nB_changed\nC_changed\nD_changed\ne\n' > dense.txt
$ git lines stage dense.txt:-2,2,-4,4
Error: git apply failed: error: patch failed: dense.txt:2
error: dense.txt: patch does not apply

This is the case the existing // Case 3: Mixed ... For now, keep as single hunk - more complex splitting could be added later comment in into_hunks flagged.

Fix

FilteredContent.additions now preserves the new-line position of each addition (Vec<(u32, String)> instead of Vec<String>), letting into_hunks split the mixed (deletions + additions) case into one hunk per contiguous deletion group, paired by index with the corresponding contiguous addition group. The split applies only when group counts match and no no-newline state is involved; asymmetric selections fall back to the previous single-hunk behavior.

After the fix:

$ git lines stage dense.txt:-2,2,-4,4
Staged:
dense.txt:
  -2:	b
  +2:	B_changed

  -4:	d
  +4:	D_changed

$ git diff --cached
@@ -1,5 +1,5 @@
 a
-b
+B_changed
 c
-d
+D_changed
 e

Test plan

• Adds e2e test replacement::non_contiguous_in_dense_hunk (3.10) covering the bug
• Updates the unit tests asserting on FilteredContent.additions for the new tuple type
• Full suite green: 83 lib + 48 e2e + 7 doc-tests = 138 tests, 0 failures
• Manual smoke test reproducing the original failure now passes

🤖 Generated with Claude Code