PERF: make RangeIndex iterate over ._range

[topper-123]

Minor performance issue.

By adding a custom __iter__ method to RangeIndex, we partly avoid needing to create/cache the expensive _data attribute and partly it's just faster to iterate over a range than a ndarray:

>>> idx = pd.RangeIndex(100_000)
>>> %%timeit
... for _ in idx:
...     pass
10.9 ms ± 74.7 µs per loop  # master
6.11 ms ± 48.8 µs per loop  # this PR
>>> "_data" in idx._cache
True  # master
False  # this PR

xref #35432, #26565.

[jreback]

cool, can you add an asv which hit this (or do we have enough coverage)?

[topper-123]

Ok, i’ve added ASVs.

[jreback]

thanks @topper-123