ci: partition crate-checks into 3 parallel jobs

[decofe]

Split crate-checks into 3 parallel matrix jobs using cargo hack check --workspace --partition M/N, matching the pattern used by reth.

Motivation

crate-checks runs cargo hack check sequentially across all 36 workspace crates. On fork PRs (which get no sccache or rust-cache), this takes ~16 min. Partitioning into 3 parallel jobs cuts wall-clock time by ~3x.

Changes

• Split into 3 matrix partitions via cargo hack check --workspace --partition 1/3
• Add fail-fast: false so one failing partition doesn't cancel the others
• Add cache-on-failure: true to rust-cache so failed runs still populate cache
• Add partition-specific rust-cache key to avoid save races between matrix legs

Prompted by: zerosnacks