Fix gramcover gauge dependence for coupled W

[timholy]

A nondiagonal W can couple distinct support components, which gramcover merges into a group. The bound over a group is built from the per-block sums M[p,q], each weighted by sqrt(M[p,p]/M[q,q]), rather than from a single uniform total over the group. Every term then scales as γ[p]^2 under the componentwise gauge a -> γ*a, b -> b/γ, so s depends on (a, b) only through the products a[i]*b[j], as it already did for diagonal W.

Where a component's own block vanishes, its weight propagates along the coupling as M[p,q]^2/M[q,q]. A group in which every block vanishes admits no such weight: the surviving data fixes each product s[j]*s[k] across two components but not how it divides between them, and the group takes the uniform total.

Assisted-by: Claude Opus 4.8 noreply@anthropic.com

[codecov]

Codecov Report

❌ Patch coverage is 97.82609% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 98.74%. Comparing base (e8d8379) to head (bd97f42).

Files with missing lines	Patch %	Lines
src/gram_covers.jl	97.82%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #37      +/-   ##
==========================================
- Coverage   98.78%   98.74%   -0.05%     
==========================================
  Files          14       14              
  Lines        2306     2383      +77     
==========================================
+ Hits         2278     2353      +75     
- Misses         28       30       +2     

☔ View full report in Codecov by Harness.
📢 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.