Testing: EvidenceExtractor abbreviation-aware splitting has no test coverage

[mrveiss]

Problem

PR #2192 (#2170) added abbreviation-aware sentence splitting to EvidenceExtractor._split_sentences() with a _ABBREV_RE pattern that merges fragments after common abbreviations (Dr., U.S.A., etc.).

However, no tests were added for the new abbreviation handling. The existing TestSplitSentences class in evidence_extractor_test.py only tests basic delimiters (period, question mark, exclamation mark) and doesn't verify:

• "Dr. Smith diagnosed the issue." stays as one sentence
• "The U.S.A. deployed Redis." stays as one sentence
• Mixed abbreviations with real sentence breaks

Discovered During

Post-implementation review of #2170 fix.

Suggested Fix

Add 2-3 test cases to TestSplitSentences in evidence_extractor_test.py:

def test_abbreviation_not_split(self):
    parts = extractor._split_sentences("Dr. Smith said hello. Then left.")
    assert len(parts) == 2
    assert parts[0] == "Dr. Smith said hello."

Impact

Low — the fix works correctly but lacks regression test coverage.