[Barbara Gittings] Docs generation fails if a database/schema name is numeric-looking

[drewbanin]

Describe the bug

Tagging this one with the BigQuery label because that's where I've seen it, but it could conceivably happen on any plugin.

If a database or schema name is numeric, then Agate's type inference will make the database_name or schema_name column (unique to a given agate table) numeric during catalog generation. This causes some calls to database_name.lower() or schema_name.lower() to fail with an error:

encountered an error while generating catalog: 'decimal.Decimal' object has no attribute 'lower'

I think I narrowed this down to _relations_filter_schemas in the base impl.py, but it's possible that there are other places where we assume the database name is going to be a string.

Beyond this first issue, Agate will also be mad about unioning together different agate tables that were fetched in parallel if one of those columns is string-y and the other is number-y in nature.

dbt should ensure that all of the known-string columns are returned as strings (and not numbers, if they look numeric) when fetching information from the system catalog.

Steps To Reproduce

Use the profile:

    dev:
      type: bigquery
      method: oauth
      project: '{project name}'
      dataset: "1991"
      threads: 1

Then run:

dbt run
dbt docs generate

System information

Which database are you using dbt with?

• postgres
• redshift
• bigquery
• snowflake
• other (specify: ____________)

The output of dbt --version:

0.16.0-rc2

[beckjake]

Fixed in #2208