Make data frame selection return row numbers, not pandas index value

[jcheng5]

Fixes #676

The code in js/dataframe/index.tsx uses the row number ([0, 1, 2, ...]) as the row.id and tag's data-key attribute. That's fine until it's time to send the selection back to Shiny, at which time we need to convert from the row number to the entry in df.index.

Actually, I suppose it's a question worth asking: what would you expect input.grid_selected_rows() to return: row numbers (to be used with df.iloc) or Pandas index values (to be used with df.loc)? I guess either is defensible? (cc @nealrichardson @machow @chendaniely)

---

Update 2023-08-15 13:32-700: Before this PR, the code was just wrong--using a row's row number to look up the keyToIndex map, which was actually not key -> index (where "index" is referring to the pandas index) but str(index) -> index. Hence the None being returned from the repro case in #676 (there were row numbers that did not exist in the pandas index). In other words, the only way the reported selection would be correct is if the pandas index is the set of integers 0..n (which, to be fair, is probably the case with most pandas data frames, as it's the default).

[jcheng5]

@andrie would love your opinion on 👆, it looked like your reprex had df["col"][input.grid_selected_rows()] which implies row numbers I think.

[jcheng5]

Actually maybe using Pandas index values is not that realistic. They can theoretically be made up of almost any kind of Python object, can't they? I'm not sure we can successfully round-trip e.g. Python datetime values through JSON, and if they come back wrong, they won't be usable with df.loc.

[andrie]

@andrie would love your opinion on 👆, it looked like your reprex had df["col"][input.grid_selected_rows()] which implies row numbers I think.

I don't have a strong opinion on this. I adapted the documented example at https://shinylive.io/py/examples/#interactively-excluding-data and tried to make minimal modifications:

    @reactive.Calc
    def filtered_df():
        selected_idx = list(req(input.summary_data_selected_rows()))
        countries = summary_df["country"][selected_idx]
        # Filter data for selected countries
        return df[df["country"].isin(countries)]

[jcheng5]

After sleeping on it, I think row numbers are the way to go. They're also less pandas-specific, if we want more native support of, I don't know, DuckDB or Ibis in the future.

[wch]

Looks good, but please see my comment about CoW operations when working with Pandas data frames.

[jcheng5]

Rebased against master--had to do this interactively because the e2e folder changed to tests/e2e on main since this PR started.