feat(api): improve efficiency of __dataframe__ protocol

[jcrist]

This improves the efficiency and usability of the __dataframe__ protocol in a few ways:

• It avoids executing the wrapped query for operations where that's not strictly required. This allows fetching things like column names and types, or subselecting columns to occur without executing the query first.
• Given a single __dataframe__() output (an IbisDataFrame instance), the wrapped query will execute at most once for all methods called. Once executed, the resulting pyarrow.Table is stored on the IbisDataFrame instance to be reused by subsequent access.
• If column subselection is done on an already executed IbisDataFrame, no computation results.

Given these improvements, an ibis.Table passed to altair (dev) now results in only one execution, rather than potentially several executions.

[jcrist]

Thanks for the review, I believe this should be good-to-go now.

[cpcloud]

Excellent work.

Excited to see this unblock some sweet integrations!