fix(snowflake): do not quote parameter names in ALTER SESSION

[DG47]

Description

Fixes #5666.

Snowflake session parameters (TIMEZONE, QUERY_TAG, ...) are keywords, not identifiers, so ALTER SESSION SET "TIMEZONE" = 'UTC' is a syntax error. sqlglot parses the parameter names in ALTER SESSION SET/UNSET as identifiers (Column(Identifier) / Identifier), so a model pre/post statement such as

ALTER SESSION SET TIMEZONE = 'UTC';

was quoted twice: once by the renderer's normalize-and-quote step and again by EngineAdapter._to_sql. It reached Snowflake as ALTER SESSION SET "TIMEZONE" = 'UTC'.

This adds a small parser override in sqlmesh/core/dialect.py, following the same pattern as the existing T-SQL ALTER COLUMN override. After sqlglot parses ALTER SESSION, unquoted parameter names become exp.Var, so they are neither normalized nor quoted. Values are left alone (literals, macros such as @tag, etc.), and so are parameter names the user quoted explicitly.

The SQL produced for the unrendered statement doesn't change (TIMEZONE = 'UTC' both before and after), so model data and metadata hashes stay the same and existing snapshots aren't affected. I checked this on a sample model.

An earlier attempt (#5670) turned quoting off for every Alter/Set statement in the Snowflake adapter and also special-cased the renderer. Its author closed it. This change only touches ALTER SESSION parameter names and lives at the parse level, so the renderer and the engine adapter both pick it up without engine-specific code in either.

Test Plan

• tests/core/test_dialect.py::test_snowflake_alter_session_parameters_are_not_quoted: parse/generate with identify=True for SET (with a macro value), UNSET, and an explicitly quoted name, plus rendered model pre/post statements.
• tests/core/engine_adapter/test_snowflake.py::test_alter_session_parameters_are_not_quoted: SQL executed by the mocked Snowflake adapter.
• Both tests fail on main (ALTER SESSION SET "TIMEZONE" = 'UTC') and pass with the fix.
• pytest -n 8 -m "fast and not web and not github and not dbt and not jupyter and not cicdonly and not isolated": 2436 passed. 6 failures in tests/core/test_connection_config.py (pyodbc) are unrelated and come from a missing local unixODBC library; they fail the same way on main.
• pytest -m "dbt and fast" tests/dbt: 88 passed.
• SKIP=prettier,eslint pre-commit run --all-files (ruff, ruff-format, mypy, migrations): passed. The prettier/eslint hooks were skipped because no frontend files changed.

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) (I ran the core fast and dbt fast subsets listed above, not the whole target)
• My commits are signed off (git commit -s) per the DCO

[Vamsi-klu]

Thanks, doing this at parse time is a lot cleaner than the adapter + renderer route from #5670. I double checked the hash claim: a Snowflake model with ALTER SESSION SET TIMEZONE = 'UTC', query_tag = 'x' as a pre statement has the same data and metadata hash on main and on this branch, only the rendered SQL changes. One scoping nit inline, looks good otherwise.

[Vamsi-klu]

On sqlglot 30.8 only the Snowflake parser sends ALTER SESSION to _parse_alter_session, Oracle and Databricks still end up as Command. So it's Snowflake only in practice, but overriding the base Parser makes it read as global. Could this be _override(Snowflake.Parser, _parse_alter_session) like the _parse_table_parts one above? Keeps the scope obvious and won't quietly pick up other dialects if sqlglot routes them here later.

[DG47]

Good call, done in 637af55: the override is now registered on Snowflake.Parser next to _parse_table_parts, and the function is annotated accordingly. I checked that Databricks and Oracle still come out as Command with the original text untouched.

[Vamsi-klu]

FYI, no change needed: since these are Vars now they don't get normalized, so query_tag stays lowercase where main uppercased it. Snowflake doesn't care about case for parameter names so it's harmless, but maybe worth a line in the comment above so nobody "fixes" it later.

[DG47]

Added a sentence to the comment in the same commit explaining that vars skip case normalization, that query_tag therefore stays lowercase, and that Snowflake treats parameter names case-insensitively so this is intentional.