C++ refactoring: layout.completely_flatten should not concatenate (performance issue).

[jpivarski]

This was raised on Slack by @martindurant and @douglasdavis.

Before:

In [1]: import awkward._v2 as ak, numpy as np

In [2]: x = np.ones((100, 100, 100))

In [3]: arr = ak.Array(x)

In [4]: %timeit ak.sum(arr, axis=None)
813 µs ± 6.08 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

In [5]: %timeit x.sum()
332 µs ± 5.38 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

After:

In [1]: import awkward._v2 as ak, numpy as np

In [2]: x = np.ones((100, 100, 100))

In [3]: arr = ak.Array(x)

In [4]: %timeit ak.sum(arr, axis=None)
317 µs ± 20.4 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

In [5]: %timeit x.sum()
295 µs ± 8.65 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

It removes an unnecessary concatenation.

[codecov]

Codecov Report

Merging #1245 (1e45797) into main (ab65e20) will increase coverage by 0.25%.
The diff coverage is 66.97%.

Impacted Files	Coverage Δ
src/awkward/_v2/_util.py	82.24% <ø> (ø)
src/awkward/_v2/contents/recordarray.py	80.60% <ø> (+0.86%)	⬆️
src/awkward/_v2/operations/reducers/ak_argmin.py	55.00% <0.00%> (-13.75%)	⬇️
src/awkward/_v2/operations/describe/ak_type.py	45.71% <38.70%> (-34.29%)	⬇️
src/awkward/_v2/operations/reducers/ak_all.py	87.50% <80.00%> (-4.17%)	⬇️
src/awkward/_v2/operations/reducers/ak_any.py	87.50% <80.00%> (-4.17%)	⬇️
src/awkward/_v2/operations/reducers/ak_count.py	87.50% <80.00%> (-4.17%)	⬇️
...wkward/_v2/operations/reducers/ak_count_nonzero.py	87.50% <80.00%> (-4.17%)	⬇️
src/awkward/_v2/operations/reducers/ak_prod.py	87.50% <80.00%> (-4.17%)	⬇️
src/awkward/_v2/operations/reducers/ak_sum.py	93.75% <80.00%> (-6.25%)	⬇️
... and 22 more