Use total-order sensitivity for high-dimensional experiments

[saitcakmak]

Summary:
Switch InsightsAnalysis to use total-order Sobol sensitivity analysis
instead of second-order when the experiment has more than 25 parameters.

Second-order sensitivity computes pairwise interaction effects at O(p^2) cost,
which becomes expensive for high-dimensional search spaces. Total-order captures
each parameter's overall importance including all interactions at O(p) cost.

For a 93-parameter GAIN experiment (GAIN_35259), total-order sensitivity is
2.7x faster than second-order (16s vs 43s per metric, with a pre-fitted adapter).

The threshold of 25 parameters preserves second-order analysis (and contour
plots showing pairwise interactions) for lower-dimensional experiments where
the O(p^2) cost is manageable.

Reviewed By: mpolson64

Differential Revision: D98506213

[meta-codesync]

@saitcakmak has exported this pull request. If you are a Meta employee, you can view the originating Diff in D98506213.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 96.41%. Comparing base (60600e0) to head (6966610).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #5115   +/-   ##
=======================================
  Coverage   96.41%   96.41%           
=======================================
  Files         613      613           
  Lines       68166    68195   +29     
=======================================
+ Hits        65720    65749   +29     
  Misses       2446     2446           

☔ 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.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[meta-codesync]

This pull request has been merged in f5976b8.