feat(python): Support DataFrame init from raw SQLAlchemy rows

[alexander-beedie]

Closes #19816.

• With a trivial extension to the existing Sequence checks we can support direct DataFrame/Series init from a list of SQLAlchemy Row objects, as they intentionally mimic namedtuple in various ways¹.

• This allows for a very small (low single-digit percentage) speedup ingesting SQLAlchemy results via read_database, so I've taken advantage of that (against a 50000x10 test dataset I was seeing ~1-2% gains - but it's not 0%, so why not...😅)

Footnotes

1. sqlalchemy.engine.Row ↩

[roganjoshp]

Just to be clear, now that you've measured it as being marginally faster (I never anticipated that), I want to state that my primary goal is to decouple the query from the dataframe construction (be that pandas or polars) - I don't like executing SQL through yet another layer that may change or not be able to take advantage of new developments. I also do other things with query results than just making dataframes.

My motivation here is not just speed gains.

I'm amazed by your fast turn-around on this PR. Thank you.

[alexander-beedie]

I don't like executing SQL through yet another layer that may change or not be able to take advantage of new developments.

On the flip side - you may miss out on optimisations we are able to make; for instance, if you use duckdb_engine for SQLAlchemy-mediated access to DuckDB, we recognise that and fetch Arrow data from the Cursor instead of fetching rows, making for a significant speedup initialising a DataFrame (you could do this yourself of course, but not everyone is aware that it's an option).

Still, allowing you more control over your own init/queries/etc if you want it is a good thing, and it sounds like you have a variety of such use-cases, so happy to enable it.

My motivation here is not just speed gains.

That's probably a good thing as the speed gain (inside read_database) is only 1-2%, so it's barely above the threshold of noise (we can skip an internal [tuple(row) for row in result] comprehension now - that's about it) ;)

I'm amazed by your fast turn-around on this PR. Thank you.

😎👍