PERF: Improve performance of cut with IntervalIndex bins

[jschendel]

• closes PERF: Avoid materializing entire IntervalIndex when using cut #27668
• benchmarks added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

ASV output:

       before           after         ratio
     [143bc34a]       [d531e7b0]
     <master>         <perf-cut-ii>
-            198M             122M     0.62  reshape.Cut.peakmem_cut_interval(1000)
-            197M             122M     0.62  reshape.Cut.peakmem_cut_interval(4)
-            197M             122M     0.62  reshape.Cut.peakmem_cut_interval(10)
-      2.11±0.02s          910±2ms     0.43  reshape.Cut.time_cut_interval(1000)
-      1.95±0.01s          761±2ms     0.39  reshape.Cut.time_cut_interval(4)
-      1.95±0.03s         763±10ms     0.39  reshape.Cut.time_cut_interval(10)

SOME BENCHMARKS HAVE CHANGED SIGNIFICANTLY.
PERFORMANCE INCREASED.

[jreback]

very nice @jschendel