ci: shard windows rust tests across 2 parallel jobs via nextest partitioning

[Ziinc]

Summary

• The windows-latest / rust / test job's cargo nextest run step alone reports 937.090s (~15.6 min) of actual test execution out of the job's ~22 minute total — the tests are dominated by jj/git subprocess spawn overhead (individual tests taking 15-49s each), not CPU-bound work.
• Splits that single matrix entry into two parallel entries using cargo nextest run --partition count:N/2, so each shard runs roughly half the tests concurrently instead of serially in one job.
• Job names now show the partition, e.g. rust / test / windows-latest (1/2).
• Reduced timeout-minutes from 45 to 30 per shard to reflect the expected halved runtime, while leaving headroom.

Why this approach

cargo nextest's built-in partitioning (--partition count:N/M) deterministically splits the test binary's test list across shards without needing to hand-maintain a test allowlist per shard, and it composes cleanly with the existing Swatinem/rust-cache step (both shards share the same cache key since they compile identical test binaries — only which tests run differs).

Test plan

• Confirm both rust / test / windows-latest (1/2) and rust / test / windows-latest (2/2) jobs pass in this PR's CI run.
• Confirm total test count across both shards' summaries adds up to the full suite (663 as of the baseline run) with no duplicates or gaps.
• Compare wall time of each shard against the ~16 min baseline for the unsharded run.
  Generated by Claude Code