Enable ak.singletons

[ioanaif]

No description provided.

[codecov]

Codecov Report

Merging #1444 (51b6cc7) into main (edfce38) will decrease coverage by 0.02%.
The diff coverage is 61.83%.

Impacted Files	Coverage Δ
src/awkward/_v2/_connect/cuda/__init__.py	0.00% <0.00%> (ø)
src/awkward/_v2/_connect/jax/__init__.py	0.00% <0.00%> (ø)
src/awkward/_v2/contents/recordarray.py	82.26% <ø> (+0.76%)	⬆️
src/awkward/_v2/contents/unionarray.py	86.50% <ø> (ø)
src/awkward/_v2/operations/convert/ak_from_cupy.py	50.00% <0.00%> (+23.33%)	⬆️
src/awkward/_v2/operations/convert/ak_from_iter.py	93.75% <ø> (ø)
src/awkward/_v2/operations/convert/ak_from_jax.py	50.00% <0.00%> (-25.00%)	⬇️
...wkward/_v2/operations/convert/ak_to_arrow_table.py	100.00% <ø> (ø)
src/awkward/_v2/operations/convert/ak_to_cupy.py	33.33% <0.00%> (+23.95%)	⬆️
src/awkward/_v2/operations/convert/ak_to_jax.py	33.33% <0.00%> (-41.67%)	⬇️
... and 74 more

[jpivarski]

To implement an axis, maybe use :, :, :, np.newaxis (if nonnegative) or ..., np.newaxis, :, :, : (if negative) and then drop_none?

As for whether singletons should include Nones, I think it shouldn't, because otherwise it would be completely identical to the above slide (and it wouldn't be doing what singletons was invented to do: to switch between a token-based and a 0/1-length list-based representation of missing data).

[jpivarski]

drop_none is #832.

[jpivarski]

Approving, but there needs to be another PR that adds an axis parameter to the new singletons. That can be implemented another PR, especially if you're going the route of doing it through drop_none.