feat: rewrite arrow_to_rdf and arrow_to_rseries_result

[eitsupi]

Fix #725, part of #677

After rewriting the contents of the functions with reference to Python Polars, the bug #725 seems to have been fixed.
(I'm not sure where the bug was occurring after all)

Other improvements:

• as_polars_df for RecordBatch now works.
• as_polars_series for Array and ChunkedArray are added.

These changes have completely ported pl$from_arrow's functionality to as_polars_*, so we can deprecate pl$from_arrow in a follow-up PR.

[etiennebacher]

Is there also a way to add a test reproducing #725 or would that be too hard to add in unit test?

[eitsupi]

Is there also a way to add a test reproducing #725 or would that be too hard to add in unit test?

Of course I'll add it. Something like below (this crashes in current release version)

  da_string = arrow::Array$create(
    factor(c("x", "y", "z"))
  )

  da_large_string = da_string$cast(
    arrow::dictionary(
      index_type = arrow::uint32(),
      value_type = arrow::large_utf8()
    )
  )

  at = arrow::arrow_table(foo = da_string, bar = da_large_string)

as_polars_df.ArrowTabular(at)

[etiennebacher]

Thanks!