Chore: Fix tests and adapt parser methods for tokens size

[themisvaltinos]

The tests currently fail and this pr addresses 1. the flakey test by making it deterministic and 2. the rest of the tests when sqlglot 30.0.3 is picked up by adapting the parser methods.

For the flakey test: test_run_with_changes_and_full_refresh by clearing cache to pick up the updated file. The reason being in _cache_entry_id it uses int(max_mtime) with the file modification time that in the unfortunate time that it is invoked the same second as the dbt cli command it matches the cached entry and reuses that instead of the fresh file.

Also sqlglot 30.0.3 that includes this: (tobymao/sqlglot@f87ebe0#diff-63eb8dd82d3561bc414e3e13cf59516bd6584c5766a10072157a8a5aee4ad85f) introduced _tokens_size, a cached copy of len(self._tokens). Ssince MacroExtractor.extract() bypasses the parser path entirely by assigning self._tokens directly this left them at 0. Since _advance() uses _tokens_size now as the boundary check every call immediately returned , causing the loop to exit before reading any tokens and silently dropping all macros defined in .sql files.

One fix for this was to update _tokens_size = len(self._tokens) immediately after the direct assignment, guarded by hasattr for compatibility with older sqlglot versions that do not have this field. And in _parse_if before delegating to the statement parser to decrement _tokens_size accordingly so not to fall out of index.