Add table lookup entry point to SqlToSubstrait

[rymurr]

This uses a function to look up a table schema without a create table statement. I've intentionally kept the interface free from calcite objects so as to not export them as part of the isthmus public API.

I am not an expert at calcite so I am not sure if this is the best way to do it. Also, note, my fix adds state to the converter. Happy to adjust based on input.

[jacques-n]

Hey @rymurr, I don't think this is the right pattern. As part of validation, the validator needs to figure out resolution based on current schema versus root schema. I also think it may be the case that some of the identifiers you'd find in nested queries wouldn't be tables (not sure on that). I think you need to plugin closer to the schema apis. That being said, I'm not sure the best pattern to do so. I remember at Dremio we actually had a fork of Calcite for exactly this purpose. Maybe @laurentgo would have some advice on whether those changes have been pushed back into Calcite using some kind of pattern.

[jinfengni]

Seems to me we need CalciteSchema that dynamically loads subschema/tables on demand when the validator requests to validate an identifier either for subschema/table. The tableLookUp function should be provided to such dynamic CalciteSchema.

This pr first finds all the sql identifier in the parse tree, and then try to match/find the identifier in tableLookUp. However, an identifier abc could be either a subschema name, or table name or column name. We would not know which kind it is, unless we have validator's context.

[rymurr]

Thanks a lot for the feedback @jinfengni and @jacques-n

I have updated to use a catalog. Do you think this is a better approach?

[rymurr]

Thanks for the review @jacques-n caught a few bugs and fixed them in the most recent iteration.

[jacques-n]

This looks reasonable to me. Merging.