fix(format): keep MODEL/AUDIT/METRIC header dialect-agnostic (#5773)

[anxkhn]

Description

sqlmesh format corrupts boolean properties in the MODEL (...) header when the model dialect has no native TRUE keyword. On a T-SQL model, a property such as allow_partials true is rewritten to allow_partials (1 = 1), and SQLMesh can no longer re-parse that as a boolean, so the model file breaks. Fixes #5773.

Root cause: format_model_expressions (sqlmesh/core/dialect.py) rendered every parsed expression with the target dialect, including the SQLMesh-specific MODEL/AUDIT/METRIC header. That header is SQLMesh DDL, not standard SQL, so SQLGlot's tsql generator transpiles a boolean to (1 = 1).

This renders meta expressions (is_meta_expression -> Audit/Metric/Model) with dialect=None on both the single-meta early return and the multi-expression join path, so the header stays exactly as written, while the actual query/statement body still transpiles to the target dialect.

Test Plan

• Added a tsql case to tests/core/test_dialect.py::test_format_model_expressions asserting the MODEL header keeps allow_partials TRUE (not (1 = 1)) while the query body still transpiles (SELECT TRUE -> 1, CAST(x AS INT) -> x::INTEGER).
• Confirmed the new assertion fails on main and passes with this change.
• pytest tests/core/test_dialect.py tests/core/test_format.py green; make style (ruff + mypy) clean on both files.

AUDIT/METRIC headers share the identical is_meta_expression code path and behave the same; happy to add an AUDIT assertion too if that would be useful.

Checklist

• I have run make style and fixed any issues
• I have added tests for my changes (if applicable)
• All existing tests pass (make fast-test)
• My commits are signed off (git commit -s) per the DCO

[StuffbyYuki]

@anxkhn Thanks for this PR!

Optional: AUDIT/false boolean test cases; text_diff may deserve the same treatment in a follow-up.