chore: Critical: fix BM25 autogenerate expression normalization

[philippemnoel]

Ticket(s) Closed

• Closes #

What

• Fix BM25 expression normalization so qualifier stripping does not alter dotted text inside SQL string literals.
• Preserve pdb namespaces while removing non-pdb relation qualifiers outside literals.
• Add regression tests for dotted literals and qualifier stripping behavior.

Why

• The previous normalization path could mangle tokenizer namespaces or dotted literal content (for example regex patterns), which can produce incorrect Alembic BM25 diff behavior.

How

• Replace regex-style qualifier stripping with token-aware parsing that tracks SQL single-quoted literals.
• Strip only non-pdb qualifier prefixes outside string literals.
• Extend unit tests in tests/unit/test_alembic_unit.py to cover the relevant normalization edge cases.

Tests

• ruff check .
• python -m pytest tests/unit/test_alembic_unit.py
• python -m pytest tests/unit