Consistent Return Structure for Rolling Apply

[WillAyd]

• closes Rolling groupby should not maintain the by column in the resulting DataFrame #14013
• tests added / passed
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

With the changes made in #20949 it appears possible to fix this bug by suppressing a catch block that was internal to Rolling's _apply. I'm not sure what the purpose of this catch as it essentially allows values to bypass the applied function...Only one test broke on removal which looked wrong anyway, so I updated it as such

[jreback]

looks good

can u add a whatsnew note 0.23 ok

[codecov]

Codecov Report

Merging #20984 into master will decrease coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master   #20984      +/-   ##
==========================================
- Coverage   91.82%   91.82%   -0.01%     
==========================================
  Files         153      153              
  Lines       49490    49486       -4     
==========================================
- Hits        45443    45439       -4     
  Misses       4047     4047

Flag	Coverage Δ
#multiple	90.21% <100%> (-0.01%)	⬇️
#single	41.85% <0%> (ø)	⬆️

Impacted Files	Coverage Δ
pandas/core/window.py	96.27% <100%> (-0.02%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update b3f07b2...878e640. Read the comment docs.

[jreback]

thanks!