ci(nextest): partition unit tests into 4 parallel shards

[bug-ops]

Summary

• Add ci-partition profile to .github/nextest.toml for hash-based test partitioning
• Replace sequential test job with 4-way matrix strategy
• Partition unit tests with --partition hash:K/4 (K=1..4) for parallel execution
• Expected wall time: ~3 min per shard (vs ~10 min sequential) = ~70% improvement
• Build job (25 min) remains the pipeline bottleneck

Changes

• .github/nextest.toml: Add [profile.ci-partition] with test-threads=8
• .github/workflows/ci.yml: Replace test job with matrix:
  • Strategy: partition: ["1/4", "2/4", "3/4", "4/4"]
  • Each shard downloads the same archive and runs its hash bucket
  • JUnit reports collected from each shard

Test plan

• Verify CI workflow passes with new 4-shard matrix
• Confirm each shard runs independently and completes in <5 min
• Check that total passing test count equals sum across all shards
• Integration and coverage jobs unchanged