feat: Order observability optimizations

[ritchie46]

An operation doesn't have to maintain_order if it isn't observed. E.g.

concat().sort() could do an unordered union for instance.

.sort().groupby().sum() can remove the sort, etc.

[codecov]

Codecov Report

Attention: Patch coverage is 98.31461% with 3 lines in your changes missing coverage. Please review.

Project coverage is 78.96%. Comparing base (2f2bb92) to head (17dda41).

Files with missing lines	Patch %	Lines
...rates/polars-plan/src/plans/optimizer/set_order.rs	98.33%	2 Missing ⚠️
py-polars/polars/lazyframe/frame.py	50.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #20396      +/-   ##
==========================================
+ Coverage   78.95%   78.96%   +0.01%     
==========================================
  Files        1561     1562       +1     
  Lines      219931   220076     +145     
  Branches     2480     2480              
==========================================
+ Hits       173643   173789     +146     
+ Misses      45717    45716       -1     
  Partials      571      571              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[s-banach]

Not in front of my computer so I can’t test an example, is groupby always faster than sort.groupby eg in case of groupby a high cardinality float column where the values might be all unique?