chore: matmul cleanup

[dvdplm]

After the bi-variate poly opt (PR #576), the matmul code left was not doing very much. This PR cleans that up a bit by introducing a VdmMatrix type and implement only the math we need:

• constructor from exceptional seq
• mul from the left by a vector

On my machine the new code is ~45% faster for the sizes we use (n=13, t=4):

group                                                  matrix-main         new                speedup
-----                                                  -----------         ---                -------
matrix/production/single_sharing_vdm/n13_t4            859.7 ns            598.2 ns           1.44×
matrix/production/double_sharing_vdm/n13_t4            1715.6 ns           1166.6 ns          1.47×
matrix/production/robust_prss_transposed_vdm/n13_t4    859.3 ns            582.1 ns           1.48×

Also removes the ndarray crate from the workspace.

[github-actions]

Consolidated Tests Results 2026-05-11 - 12:59:49

Test Results

12 passed

Details

12 tests
not captured
junit-to-ctrf
build-and-test test-reporter #2056
chore: matmul cleanup #585

test-reporter: Run #2056

Tests 📝	Passed ✅	Failed ❌	Skipped ⏭️	Pending ⏳	Other ❓	Flaky 🍂	Duration ⏱️
12	12	0	0	0	0	0	not captured

🎉 All tests passed!

Tests

View All Tests

Test Name	Status	Flaky	Duration
k8s_test_crs_uniqueness	✅		31.9s
k8s_test_insecure_keygen_encrypt_and_public_decrypt	✅		2m 6s
k8s_test_insecure_keygen_encrypt_multiple_types	✅		2m 18s
k8s_test_keygen_and_crs	✅		2m 1s
k8s_test_keygen_uniqueness	✅		5m 15s
k8s_test_crs_uniqueness	✅		32.3s
k8s_test_insecure_keygen_encrypt_and_public_decrypt	✅		2m 6s
k8s_test_insecure_keygen_encrypt_multiple_types	✅		2m 20s
k8s_test_keygen_and_crs	✅		2m 2s
k8s_test_keygen_uniqueness	✅		5m 19s
k8s_test_centralized_insecure	✅		53.9s
nightly_full_gen_tests_default_k8s_centralized_sequential_crs	✅		1.7s

_{🍂 No flaky tests in this run.}

Github Test Reporter by CTRF 💚

🔄 This comment has been updated