typecheck workflow has no concurrency group, so stale runs are never cancelled

Description

typecheck.yml has no concurrency block, so pushing again to a PR leaves the previous typecheck run going instead of cancelling it. test.yml already handles this.

This is visible in the run history. For one push to the zed-acp-docs PR:

• test run 32718073842 was cancelled after 9m37s when the next push arrived, which is its concurrency group working.
• typecheck run 32718073884, from that same push, ran to completion (1m9s) and was never cancelled, then 32718860388 started on top of it.

So every extra push to a PR burns another full typecheck on a blacksmith-4vcpu-ubuntu-2404 runner for a result nobody looks at.

test.yml already has a group expression that handles the dev-branch case correctly — it keeps every run on dev so cancelled checks don't pollute the default branch history, while letting PR runs cancel stale ones. The same expression applied to typecheck.yml would fix this.

containers.yml, generate.yml and docs-locale-sync.yml also run without a concurrency group, but I left those alone since cancelling them mid-run may not be safe in the same way.

Steps to reproduce

1. Push to a PR branch twice in quick succession
2. gh run list --workflow=typecheck.yml — the earlier run is still in progress or completed rather than cancelled
3. Compare with gh run list --workflow=test.yml for the same pushes, where the earlier run is cancelled