BUG: Bug in quantile() and median() returned wrong result for non monotonic window borders

[phofl]

• closes BUG: rolling.median/quantile does not work with (some) BaseIndexer subclasses #37153
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

cc @mroeschke

Values were not added to the window again if starting point was moved back and values were not removed from the window, when ending point was moved back

Should I create functions for the deletion and addition?

[mroeschke]

Yeah would be great to create a function that median and quantile could share for the skiplist inserts and removal

[phofl]

Should we move the for the for j in range(s, e): blocks into the function too? Would be differing from the other parts, but otherwise the assignment of err is a bit long winded

[mroeschke]

Hmm if it makes the code more DRY then sure.

[jreback]

this is orthogonal to #36933 ?

[mroeschke]

this is orthogonal to #36933 ?

Yes, median and quantile are slightly different from sum/mean/var

[phofl]

We can use is_monotonic_start_end_bounds here too nevertheless. Did miss that previously. I would rename the function after #36933 is merged.

[mroeschke]

Mind merging in master? #36933 was recently pulled in (but shouldn't directly interfere)

[phofl]

Thanks for the heads up. Had to rename the method, but this was expected. I think this could be merged now.

[mroeschke]

Appears that there's another merge conflict.

[mroeschke]

Thanks @phofl