Bound MyersDiff work for large edit distances - #4450

Fixes #2345

Problem: MyersDiff allocates O(D²) path nodes, so hasSameTextualContentAs on large, very different files runs for minutes or hours.
Fix: past MAX_EDIT_DISTANCE (10,000) edits, report the region between the common prefix and suffix as a single ChangeDelta.

Measurements (JDK 25, two runs each, allocation via ThreadMXBean):

• 16k fully different lines: 5.9–6.1 s / 11,728 MB → 0.68–0.70 s / 1,154 MB (same failure message length)
• 8k fully different lines: 1.3 s / 2,937 MB → 0.66 s / 1,152 MB
• 0.1% different, 1k–16k lines: unchanged (54–86 ms, 4–8 MB)

Tests: new MyersDiff_diff_Test (fails with the limit disabled: 2 deltas instead of 1); 89 diff/textual-content tests pass; ./mvnw spotless:apply clean.

Check List:

• Fixes hasSameTextualContentAs is hanging with text file #2345
• Unit tests : YES
• Javadoc with a code example (on API only) : NA
• PR meets the contributing guidelines

🤖 Generated with Claude Code