Improvements to send_dataframe API

[jleibs]

The send_dataframe API depends on arrow metadata tags to figure out the types of each columns.

However, this means adding another user-step of converting to a metadata-preserving format and then manually applying all of the correct tags.

We should try to reduce this friction where possible

Better timeline inference

Unlike arrow/polars/datafusion dataframes, which are pure tables with uniform columns, Pandas dataframes have a concrete index, which we could always map to a timeline of the corresponding name.

Entity/Component Tagging

As for the other columns, it would still be helpful to provide some way of informing Rerun of the entity/component for each columns. where augmenting the arrow metadata may be non-trivial.

This could maybe look something like:

df_components = [rr.Position3D, rr.Colors, "user.Confidence"]

rr.send_dataframe(df, components=df_components)

where the components arrray must match the number of columns in the dataframe.

Or maybe with an object-helper similar to AnyValues:

rr.send_dataframe(rr.TaggedDataframe(df, df_components))

[Famok]

I would love to see this!

Maybe you could also just assume its a timeseries when the values are numeric when no components are given or when one component is given (but more cols) then apply that type to every column (same as radii and colors do for some archetypes)

Concerning the time column, you could solve it with:
rr.send_dataframe(df, components=df_components, time_column = "index" ) # with index as default

about the components: I'd prefer a dictionary, so that one can skip columns or they can change in order, e.g. for a df with cols = ['xyz', 'colors']:
df_components = {'xyz': rr.Position3D, 'colors': rr.Colors}
(maybe this could also cover the time column by adding {'index': rr.Time} given rr.Time exists)