Add Graded RBP metric class

[sushobhan2024]

This PR is regarding the issue #1031

• The main change is around getting an array relevance instead of binary array.
• RBP is now configurable with optional grade_field. The grades are assumed to be [0, 1] scaled.
• Added a helper method graded_rank_biased_precision to compute scores using relevance and weights.
• Added tests in existing tests for RBP

[codecov]

Codecov Report

❌ Patch coverage is 89.47368% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 88.26%. Comparing base (e5c0071) to head (11b733a).
⚠️ Report is 4 commits behind head on main.

Files with missing lines	Patch %	Lines
src/lenskit/metrics/ranking/_rbp.py	89.47%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1032      +/-   ##
==========================================
- Coverage   89.43%   88.26%   -1.18%     
==========================================
  Files         222      222              
  Lines       15356    15387      +31     
==========================================
- Hits        13734    13581     -153     
- Misses       1622     1806     +184     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[samiravaez]

Missing test items get unknown_grade with a default of 0.25, and the tests also check for that same default. Was that intentional? It doesn't seem righ to me

[sushobhan2024]

Good point. It probably makes sense to split this into two tests: one that verifies the default behavior, and another that verifies the parameter is actually being used.

[samiravaez]

Also I'm not sure if we should have a normalization option for RBP, but if it's only for experimental purposes, I think it can exceed 1 with the current default for missing test items.
To get a lower bound score I think we should consider missing items' scores as zero. If we want an upper bound we could add the residual p^k instead of assigning a fixed value.