chore: use greedy pairwise coverage for CI matrix generation - #3965

Replace the random-per-axis row generation with a greedy algorithm
that maximizes pairwise parameter coverage. For each row, 1000
random candidates are generated and the one covering the most
previously-uncovered parameter pairs is selected.

Pair scoring uses normalized weights (weight_i × weight_j) so that
production-like values (e.g. autosave=never, query_mode=extended)
are prioritized while non-default values still appear for coverage.
Candidate generation uses uniform random to ensure diverse candidates.

This achieves ~43-47% pair coverage and ~76% weighted coverage
with 5 jobs.

Use `node matrix.mjs --coverage` to see pair coverage statistics.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Looks like 1s timeout was too agressive