GH-102613: Improve performance of pathlib.Path.rglob()

[barneygale]

Stop de-duplicating results in _RecursiveWildcardSelector. A new _DoubleRecursiveWildcardSelector class is introduced which performs de-duplication, but this is used only for patterns with multiple non-adjacent ** segments, such as path.glob('**/foo/**'). By avoiding the use of a set in most cases, PurePath.__hash__() is not called, and so paths do not need to be parsed and (case-) normalised.

Also merge adjacent ** segments in patterns.

Timings:

$ ./python -m timeit -s 'from pathlib import Path; p = Path()' 'list(p.glob("**/*"))' 1 loop, best of 5: 197 msec per loop # before 2 loops, best of 5: 146 msec per loop # after --> 35% faster 

$ ./python -m timeit -s 'from pathlib import Path; p = Path()' 'list(p.glob("**/**/*"))' 1 loop, best of 5: 1.77 sec per loop # before 2 loops, best of 5: 146 msec per loop # after --> 12x faster 

$ ./python -m timeit -s 'from pathlib import Path; p = Path()' 'list(p.glob("**/*/**"))' 1 loop, best of 5: 738 msec per loop # before 1 loop, best of 5: 731 msec per loop # after --> about the same 

• Issue: Path.rglob performance issues in deeply nested directories compared to glob.glob(recursive=True) #102613