Feat: enable dbt transpilation

[treysp]

This PR allows dbt project models and tests to be written in a different SQL dialect than the target engine on which the project is being run.

A project has a default dialect - all models/tests without an explicitly configured dialect are assumed to be written in the default. In a dbt project without an explicit default dialect specified, the target engine dialect is the default.

Specify a default dialect other than the target engine in the config.py file model defaults argument

• Example: `config=sqlmesh_config(Path(file).parent, model_defaults=ModelDefaultsConfig(dialect="bigquery"))

Specify dialects for individual models/tests in the standard way:

• Configuration jinja at top of file: {{ config(dialect='bigquery') }}
• YAML configuration files in file directories
• dbt_project.yml

Limitations

• SQLMesh allows kind-specific parameters like time_column to be specified in a different dialect than the model query itself. In dbt, kind-specific parameters are parsed with the model's dialect.

[crericha]

Since we now need to handle dialect in the jinja, would it make sense to always pass in dialect here? It would simplify create_builtin_globals code.

[crericha]

Also, do we want to pass in the default dialect? Wouldn't it make more sense to pass in the model dialect during to_sqlmesh in the model?

[treysp]

We can't pass in self.default_dialect because target hasn't been set the first time this is called and we'll error.

I'm not following your second comment - where in to_sqlmesh can we modify the globals? Here?
https://github.com/TobikoData/sqlmesh/blob/b7c659e66b6eab5ec30a6f5d341937d8746de683/sqlmesh/dbt/model.py#L382

[izeigerman]

@crericha I think this is only relevant in the context of dbt projects and doesn't apply to native projects, so not sure whether we need the dialect as a top-level argument.

[crericha]

I'm not sure this is needed. Aren't we creating directly into the generic AST form and the sqlglot generator will handle transpiling for us? How is partitioned_by parsing handled for a native sqlmesh model?

[treysp]

exp.to_column(val) is parsing the raw column string from the model definition and returning an expression. If someone includes dialect-specific characters, I think it will choke.

cluster_by is also parsing column names just below and passes a dialect, so maybe it not being here already is an oversight?
https://github.com/TobikoData/sqlmesh/blob/b7c659e66b6eab5ec30a6f5d341937d8746de683/sqlmesh/dbt/model.py#L377

Either way, this initial implementation is messy, so I'll clean it up and we can remove if it's not necessary.