Support mutating BigQuery nested types without dropping

[plaflamme]

This is a feature request for the BigQuery engine.

Currently, sqlmesh supports nested types (i.e.: STRUCT) within models, e.g.:

SELECT STRUCT(foo, bar) AS my_struct FROM underlying

It also supports changes to these, e.g.:

SELECT STRUCT(foo, bar, new_field) AS my_struct FROM underlying

But because BigQuery does not support mutating nested types using SQL, the way this is implemented in sqlmesh requires dropping the column and creating a new one (which truncates the data).

The feature request is to fallback on using the BigQuery API directly for doing these mutations. Specifically, it should be possible to add columns to a nested type without dropping it.

The caveat is that adding columns to a nested type can only be done "at the end" (i.e.: STRUCT(foo, new_field, bar) is not possible, only STRUCT(foo, bar, new_field)).

Ideally, sqlmesh would allow appending fields to a STRUCT as long as the query also matches the resulting schema.