feat: Compile base & BigQuery's parser with mypyc

[VaggelisD]

This PR enables compilation for parser.py and as a PoC it also fully compiles BigQuery's parser. The point of this PR is to:

1. Ensure that we can compile the base Parser
2. Show that other parsers can remain interpreted and nested within their Dialects, maintaining backwards compatibility
3. Show that we can incrementally move each subparser to its own file for full compilation

The key changes include:

• Extracting BigQuery.Parser into its own file for compilation alongside parser.py (mypyc cannot deal with nested classes well)
• Exposing parser dicts (_FUNCTIONS, _NO_PAREN_FUNCTIONS etc.) as module-level variables so dialect subclasses can reference them without __dict__ introspection (mypyc stores class attributes as getset descriptors instead of plain dicts, so __dict__ walking doesn't work)
• Add a _pa helper to handle mypyc-compiled class attributes by reading from a temporary instance instead of __dict__ (same mypyc getset descriptor limitation)
• Fix _NO_PAREN_FUNCTIONS bug: TokenType.CURRENT_DATETIME was mapped to exp.CurrentDate instead of exp.CurrentDatetime (pre-existing bug, exposed by the _pa fallback path triggered under mypyc)
• Switch sqlglotc build dependency from upstream mypy to sqlglot-mypy via build isolation, keeping upstream mypy for type checking

Sqlglot-mypy

sqlglot-mypy is a lightweight fork of mypy that carries mypyc bug fixes needed to compile sqlglot e.g things like cross-module class attribute defaults and globals export for separate compilation. It is only used at build time (via [build-system] requires + pip build isolation) and does not affect the main venv or type checking, which continue to use upstream mypy. This is intended to be used until all the bugs that we run into SQLGlot are fixed upstream, and as a testing ground for our own mypyc experiments.

[github-actions]

SQLGlot Integration Test Results

Comparing:

• this branch (sqlglot:compile-parser-poc, sqlglot version: compile-parser-poc)
• baseline (main, sqlglot version: 29.0.2.dev29)

⚠️ Limited to dialects: bigquery, duckdb, snowflake

By Dialect

dialect	main	sqlglot:compile-parser-poc	transitions	links
bigquery -> bigquery	2592/2624 passed (98.8%)	2592/2624 passed (98.8%)	No change	full result / delta
bigquery -> duckdb	2077/2623 passed (79.2%)	2078/2623 passed (79.2%)	1 fail -> pass	full result / delta
duckdb -> duckdb	4003/4003 passed (100.0%)	4003/4003 passed (100.0%)	No change	full result / delta
snowflake -> duckdb	1617/2642 passed (61.2%)	1645/2642 passed (62.3%)	28 fail -> pass	full result / delta
snowflake -> snowflake	2863/2863 passed (100.0%)	2863/2863 passed (100.0%)	No change	full result / delta

Overall

main: 14755 total, 13152 passed (pass rate: 89.1%), sqlglot version: 29.0.2.dev29

sqlglot:compile-parser-poc: 14755 total, 13181 passed (pass rate: 89.3%), sqlglot version: compile-parser-poc

Transitions:
29 fail -> pass

[georgesittas]

Getting an initial pass in before moving on to parser.py and parsers/bigquery.py

[georgesittas]

Final set of comments from me, LGTM otherwise.