fix: uneccessary use of nplike_of

[agoose77]

In preparation for a simplification of nplike identification, this PR eliminates redundant nplike_of usages, and/or replaces them with backend_of.

In future, nplike_of will only accept array objects (not layouts, etc).

[jpivarski]

This is different from the one I've been seeing develop, and it looks pretty simple: making the calls to get nplikes more uniform in preparation for a refactoring. Looks good!

In future, nplike_of will only accept array objects (not layouts, etc).

By "array objects," do you mean buffers, like np.ndarray, cp.ndarray, or TypeTracerArray, not ak.Array? (It seems like that's what you mean.)

[agoose77]

By "array objects," do you mean buffers, like np.ndarray, cp.ndarray, or TypeTracerArray, not ak.Array? (It seems like that's what you mean.)

In an ecosystem with N array types, you'd think I'd be more precise by now ;)

Yes, I mean the nplike buffers. The rewrite of #2307 will simplify nplike_of to simply cache the nplike which returns True for is_own_array_type, and make backend_of do the heavy lifting w.r.t to inspecting layouts, arraybuilder, etc.

[codecov]

Codecov Report

Merging #2315 (4188006) into main (df1eec6) will decrease coverage by 0.02%.
The diff coverage is 95.23%.

Additional details and impacted files

Impacted Files	Coverage Δ
src/awkward/behaviors/string.py	76.82% <75.00%> (-0.16%)	⬇️
src/awkward/_broadcasting.py	89.23% <100.00%> (-0.03%)	⬇️
src/awkward/_connect/numpy.py	92.99% <100.00%> (ø)
src/awkward/_errors.py	80.82% <100.00%> (+0.30%)	⬆️
src/awkward/_util.py	75.55% <100.00%> (-0.05%)	⬇️
src/awkward/operations/ak_broadcast_arrays.py	100.00% <100.00%> (ø)
src/awkward/operations/ak_concatenate.py	96.29% <100.00%> (ø)
src/awkward/operations/ak_full_like.py	98.21% <100.00%> (-0.07%)	⬇️
src/awkward/operations/ak_linear_fit.py	89.09% <100.00%> (+1.59%)	⬆️
src/awkward/operations/ak_mask.py	95.45% <100.00%> (-0.20%)	⬇️
... and 1 more

... and 1 file with indirect coverage changes