[blegat]

For a PSD block whose constraint matrices are all rank-one, store the m × d factors instead of the sparse d² × m Jacobian. jtprod is computed as U' * Diagonal(w .* y) * U and jprod as rowwise dots of U * V with U, both with dense BLAS products.

H and the Schur matvec are bitwise identical to before on all three instances. All seven relevant test files pass, and the existing rank-one tests now assert that a RankOneJacobian is used.

	bl/rank1	bl/rank1_jacobian
r1, d = 40, m = 800: setup	84 ms / 74 MB	1.4 ms / 5.9 MB
unsafe_jtprod	0.60 ms	0.055 ms
add_jprod!	1.05 ms	0.053 ms
r1, d = 300, m = 100: setup	340 ms / 484 MB	0.25 ms / 3.1 MB
unsafe_jtprod	6.5 ms	0.28 ms
add_jprod!	8.1 ms	0.29 ms
sparse SDP: all calls	unchanged	unchanged

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.55%. Comparing base (cc52525) to head (5a6f983).

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #93      +/-   ##
==========================================
+ Coverage   99.54%   99.55%   +0.01%     
==========================================
  Files          29       29              
  Lines        1745     1792      +47     
==========================================
+ Hits         1737     1784      +47     
  Misses          8        8              

☔ 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.