feat: Implement string, boolean and binary dtype in top_k

[CanglongCl]

This allows more dtype to use top_k which only supported numeric dtypes before.

[codecov]

Codecov Report

Attention: Patch coverage is 56.83453% with 60 lines in your changes are missing coverage. Please review.

Project coverage is 81.13%. Comparing base (31df06d) to head (0bc66cc).
Report is 14 commits behind head on main.

Files	Patch %	Lines
crates/polars-ops/src/chunked_array/top_k.rs	64.22%	44 Missing ⚠️
crates/polars-arrow/src/array/boolean/mutable.rs	0.00%	16 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #15488      +/-   ##
==========================================
- Coverage   81.14%   81.13%   -0.01%     
==========================================
  Files        1362     1362              
  Lines      174951   175115     +164     
  Branches     2533     2532       -1     
==========================================
+ Hits       141960   142083     +123     
- Misses      32508    32551      +43     
+ Partials      483      481       -2     

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

[stinodego]

Thanks for the PR, but this is not a valid solution. The whole point of top_k is that it is more efficient than sort followed by head.

If we want to support more datatypes in top_k, we should implement the algorithm for those datatypes. Booleans can have a dedicated fast implementation because we know there are only two possible values. For strings, I don't know the correct approach.

[CanglongCl]

Thanks for the PR, but this is not a valid solution. The whole point of top_k is that it is more efficient than sort followed by head.

If we want to support more datatypes in top_k, we should implement the algorithm for those datatypes. Booleans can have a dedicated fast implementation because we know there are only two possible values. For strings, I don't know the correct approach.

Thanks for review. I've rewrite the pr. binary and string now using the same algorithm with numeric types.

[ritchie46]

Almost there. Only one remark on the string path.

[ritchie46]

Alright. Going in. Thank you!