ci(unit-tests): split rust-and-cairo + add Swatinem/rust-cache

[saroupille]

Why

The combined rust-and-cairo job started clipping cold runs once the wallet test suite grew. PR #25 added the concurrent-fetch + parallel-decrypt machinery; the resulting +16s on tzel-wallet-app --lib pushed total wallclock past ~9 min and hit a runner shutdown signal mid-Cairo on the first post-merge main run (SHA 7e48045):

##[error]The runner has received a shutdown signal. This can happen when 
the runner service is stopped, or a manually started runner is canceled.

The Rust unit tests had completed successfully (122 wallet-app tests in 57s, all other crates green). Only the trailing Cairo step got reaped. The org-level cap appears to be a wallclock budget, not a workflow timeout — timeout-minutes isn't set anywhere in the file.

This pattern was also hit transiently on PR #24 before its CI-runtime trim landed; PR #25 added enough additional test work to make it consistent.

What

Two changes in .github/workflows/unit-tests.yml, no code changes:

1. Split rust-and-cairo into two parallel jobs — rust-tests and cairo-tests. Each gets its own runner / wallclock budget. Cairo no longer waits on Rust completion; failures surface independently.

2. Add Swatinem/rust-cache@v2.7.8 (SHA-pinned per the workflow's third-party-action convention) to both Rust-touching jobs (rust-tests and ocaml's cross-impl interop step). Cargo registry + target/ amortise across runs; cache key bumps on Cargo.lock change. Warm runs should drop from ~9 min cold to ~3-4 min.

The ocaml job is unchanged structurally — just gains the cargo cache for its Rust cross-impl test.

Test plan

• This PR's own CI is the primary signal. First run is cold (no cache hit yet) — should still complete inside the per-job budget since Rust and Cairo aren't sequential anymore.
• Subsequent main runs warm-cache: expect ~3-4 min wallclock per job vs ~9 min today.
• No semantic test change. All existing tests still run, just split across two runners.

🤖 Generated with Claude Code