Bump SQLGlot minimum version to 30.14

[dvasdekis]

Summary

Update the SQLGlot dependency from ~=30.8.0 to >=30.14.0,<31.0.0 so SQLMesh can receive the BigQuery-to-DuckDB transpilation fixes released in SQLGlot 30.14.0 while staying within the current major version.

Compatibility

The dependency update also requires compatibility changes for SQLMesh parser and generator overrides:

• Accept and forward SQLGlot's alias_tokens argument in _parse_join.
• Accept and forward SQLGlot 30.17's parse_function_unit argument in _parse_interval_span.
• Avoid duplicating T-SQL nullability when SQLGlot 30.17 renders it through _alter_column_null_constraint_sql.
• Preserve raw Jinja source during macro extraction and optimized-query cache key generation.
• Restore strict time-format compatibility, dialect-specific clustering generation, DuckDB safe division, and duplicate Python model detection.
• Retain regression coverage for collated types, T-SQL ALTER COLUMN nullability, and quoted Jinja macro arguments.

Fixes #5948

Validation

All local validation is green at commit 2c77676c with SQLGlot 30.17.0:

• Main fast/slow suite (pytest -n auto -q -m "(fast or slow) and not cicdonly"): 3334 passed, 20 skipped.
• Isolated phase (pytest -q -m isolated): 3 passed.
• Registry isolation (pytest -q -m registry_isolation): 1 passed.
• Dialect isolation (pytest -q -m dialect_isolated): 163 passed.
• SQLGlot 30.14.0 dialect compatibility suite: 163 passed.
• Dialect and macro tests: 305 passed.
• Ruff formatting, Ruff lint, and git diff --check: passed.

The main suite includes the Spark tests with Java 17 configured locally. The GitHub status checks are separate from this local validation and are still pending.