First attempt at genericizing data source

[jcheng5]

Do not merge this PR. Already accounted for in #19, #20 via c6a24a7

Close this PR when #20 is merged.

---

This PR is intended to make querychat extensible beyond in-memory Pandas dataframes, including but not limited to:

1. SQLite
2. PostgreSQL
3. Snowflake
4. Databricks
5. Redshift?
6. Bigquery?

It introduces a protocol called DataSource, which so far has two concrete implementations: DataFrameSource and SQLiteSource.

TODO

• Use narwhals instead of Pandas directly
• Have options to limit the number of rows retrieved, or perhaps retrieve in chunks or a cursor, so big databases won't blow up your Shiny app
• PostgreSQL
• Snowflake
• Databricks

[jcheng5]

I've decided to focus on SQLAlchemy, which gives us all the features we want for all the databases we care about, as far as I can tell. Now you can do this:

pip install SQLAlchemy 

from sqlalchemy import create_engine

querychat_config = querychat.init(
    SQLAlchemySource(
        create_engine(f"sqlite:///{Path(__file__).parent / 'titanic.sqlite3'}"),
        "titanic",
    ),
    greeting=greeting,
    data_description=data_desc,
)

This may simplify further in the future so you only need to pass the connection string and not call create_engine at all, but Claude 3.5 Sonnet feels strongly that this would be surprising for SQLAlchemy users 😄

@blairj09 See https://docs.snowflake.com/en/developer-guide/python-connector/sqlalchemy for doing this with Snowflake.

[jcheng5]

We've moved development to generic-datasource-improvements since James is presenting off of generic-datasource for Snowflake and Databricks conferences these next couple of weeks.

[chendaniely]

closing since #20 is merged