perf: Add sampling to new-streaming equi join to decide between build/probe side

[orlp]

The new streaming engine's join will now buffer at least POLARS_JOIN_SAMPLE_LIMIT rows (default 10 million) from both sides of the join before deciding which side will become a build or probe side.

[codecov]

Codecov Report

Attention: Patch coverage is 0% with 326 lines in your changes missing coverage. Please review.

Project coverage is 79.78%. Comparing base (e8e0295) to head (ec1d5eb).
Report is 51 commits behind head on main.

Files with missing lines	Patch %	Lines
crates/polars-stream/src/nodes/joins/equi_join.rs	0.00%	309 Missing ⚠️
crates/polars-expr/src/hash_keys.rs	0.00%	17 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #21197      +/-   ##
==========================================
- Coverage   79.88%   79.78%   -0.11%     
==========================================
  Files        1593     1593              
  Lines      227649   228096     +447     
  Branches     2600     2600              
==========================================
+ Hits       181860   181980     +120     
- Misses      45192    45519     +327     
  Partials      597      597              

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