fix: set dtype in full_like

[agoose77]

This PR directly sets dtype in the nplike.full_like call, which I think is all we need to do here.

Fixes #2250

[agoose77]

@jpivarski I also added support for EmptyArray if a dtype is passed. Unlike ak.values_astype, I didn't add a including_unknown flag because this seems like a bugfix and a sensible choice. Is there a reason that we added including_unknown to values_astype besides backwards compatibility? Would you like me to do the same here?

[jpivarski]

The reason is that sometimes an unknown type fills in for a NumpyArray, and sometimes it fills in for something else.

In a situation like this:

>>> partition_1 = ak.from_json("[[1.1, 2.2], [3.3]]")
>>> partition_1.type.show()
2 * var * float64
>>> 
>>> partition_2 = ak.from_json("[[]]")
>>> partition_2.type.show()
1 * var * unknown
>>> 
>>> partition_3 = ak.from_json("[]")
>>> partition_3.type.show()
0 * unknown

you wouldn't want to fill in the unknown types as a byproduct of turning the floats into integers or something. You'd want these unknowns to stay unknown until concatenation.

But there are other situations in which you want unknown to become numeric, like np.array([]).

So I think it would be a good idea for both values_astype and full_like (with dtype) to have a boolean including_unknown argument.

[agoose77]

But there are other situations in which you want unknown to become numeric

That's a good point, and links to our conversations yesterday!

[codecov]

Codecov Report

Merging #2251 (7aebbaa) into main (dc8f334) will decrease coverage by 0.02%.
The diff coverage is 100.00%.

Additional details and impacted files

Impacted Files	Coverage Δ
src/awkward/operations/ak_full_like.py	98.27% <100.00%> (+0.09%)	⬆️
src/awkward/operations/ak_ones_like.py	91.66% <100.00%> (ø)
src/awkward/operations/ak_zeros_like.py	93.33% <100.00%> (ø)
src/awkward/operations/ak_strings_astype.py	96.15% <0.00%> (-3.85%)	⬇️
src/awkward/contents/unionarray.py	83.44% <0.00%> (-0.68%)	⬇️

[jpivarski]

I looked over this when it was in the draft stage, and just did again now. I think it's great and ready to be merged!