fix: off-by-one error in run_lengths

[agoose77]

ak.run_lengths([[3, 3, 3, 3], [3]]) does not correctly consider the inner list boundary. This PR fixes the causative off-by-one error.

The cause of this bug is that the logic that forcibly splits runs across list boundaries was ignoring one-length sublists at the end of the list. This off-by-one error meant that runs were incorrectly computed across a boundary in such cases.

[agoose77]

Hmm, this test isn't failing on main...

[codecov]

Codecov Report

Merging #2093 (cf3abfb) into main (00da8e3) will decrease coverage by 0.08%.
The diff coverage is n/a.

Additional details and impacted files

Impacted Files	Coverage Δ
src/awkward/operations/ak_run_lengths.py	90.90% <ø> (ø)
src/awkward/operations/ak_sort.py	60.00% <0.00%> (-40.00%)	⬇️
src/awkward/operations/ak_argsort.py	75.00% <0.00%> (-25.00%)	⬇️
src/awkward/operations/ak_ones_like.py	90.90% <0.00%> (-9.10%)	⬇️
src/awkward/operations/ak_zeros_like.py	92.85% <0.00%> (-7.15%)	⬇️
src/awkward/operations/ak_ravel.py	93.33% <0.00%> (-6.67%)	⬇️
src/awkward/operations/ak_isclose.py	94.44% <0.00%> (-5.56%)	⬇️
src/awkward/operations/ak_count_nonzero.py	77.27% <0.00%> (-2.73%)	⬇️
src/awkward/operations/ak_nan_to_num.py	98.03% <0.00%> (-1.97%)	⬇️
src/awkward/operations/ak_full_like.py	98.07% <0.00%> (-1.93%)	⬇️
... and 15 more

[jpivarski]

I was thinking that if this was something v1 got right and v2 (without this PR) gets wrong—which I just confirmed, using your test example—then we should be able to look at the v1 code and understand why.

Here's the relevant section of the v1 code:

awkward/src/awkward/operations/structure.py

Lines 338 to 342 in 80bbef0

	diffs = data[1:] != data[:-1]
	if isinstance(diffs, ak.highlevel.Array):
	diffs = nplike.asarray(diffs)
	if offsets is not None:
	diffs[offsets[1:-1] - 1] = True

The code here is pretty different now because of is_interior, introduced in #1795. I guess that means this PR is a bug-fix on that PR, right?

Anyway, it looks right now and I think it should be merged!

[agoose77]

@jpivarski yes, don't run git blame ;)