ENH: Improve performance for arrow dtypes in monotonic join

[phofl]

• closes #xxxx (Replace xxxx with the GitHub issue number)
• Tests added and passed if fixing a bug or adding a new feature
• All code checks passed.
• Added type annotations to new arguments/methods/functions.
• Added an entry in the latest doc/source/whatsnew/vX.X.X.rst file if fixing a bug or adding a new feature.

idx = Index(list(range(1, 1_000_000)), dtype="int64[pyarrow]")
idx2 = Index(list(range(100_000, 1_100_000)), dtype="int64[pyarrow]")
idx.union(idx2)

# main
# %timeit idx.union(idx2)
# 327 ms ± 72.6 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

# pr
# %timeit idx.union(idx2)
# 2.79 ms ± 27.5 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

[jbrockmendel]

no objection here, but eventually we ought to find a way to do this dispatch without special-casing inside the index code (i.e. implement something at the EA level)

hows is perf affected on multi-chunk pyarrow objs?

[phofl]

Arrays have 2 million entries, initial performance 380ms, on this pr

%timeit idx.union(idx2)
11.5 ms ± 477 µs per loop (mean ± std. dev. of 7 runs, 1 loop each)

2 chunks
I totally agree with your point regarding EA interface, this is a short term solution for 2.0

[phofl]

@jbrockmendel ok to merge?

[jbrockmendel]

_from_sequence?

[phofl]

good point, changed

[jbrockmendel]

LGTM