Fix(athena): Properly extend Athena dialect

[erindru]

Closes #5065

I think Athena has been broken since SQLMesh v0.198.0 were we upgraded SQLGlot to v26.32.0.

With that upgrade:

• Athena stopped extending Trino and became its own top-level Dialect
• It also defined some extra top-level Parser/Tokenizer/Generator classes that could not be reached directly from <dialect>.Parser, <dialect>.Tokenizer etc

This PR includes the extra set of top-level classes when extending the Athena dialect and adds a test that checks MODEL (...) can be roundtripped across every dialect

[erindru]

todo: revert prior to merge

[georgesittas]

Thanks for taking this on. I'm surprised this wasn't caught by any tests.

[erindru]

I'm surprised this wasn't caught by any tests.

Me too. I think it was partially working for the parser because Hive.Parser is not instantiated by the Athena dialect until runtime and it had already been patched.

It was the Generator path that was getting missed because:

• the class is created at import time before the parent class Trino.Generator gets monkeypatched
• it redefines TRANSFORMS, so the runtime changes to inject the SQLMesh transforms into the base class don't flow through

We also have a tendency to assume in the test suite that if it works for duckdb, it works for everything