Optimize `deno task check-all` performance without reducing `pre-commit` coverage

[dahlia]

We currently run deno task check-all from the Git pre-commit hook. That full validation step is important and should remain in place, but the type-checking phase currently takes long enough to make routine commits slower than they need to be.

This issue is about improving the performance of check-all, not weakening the checks it performs. In particular, we should preserve full type checking in the pre-commit path and instead investigate why the current deno check step is so expensive across the workspace.

Areas to investigate:

• Identify which packages or dependency graphs dominate the total deno check time
• Determine whether the current workspace-wide invocation causes redundant type checking
• Look for opportunities to improve caching, task structure, or command composition
• Review whether generated artifacts or package boundaries are causing unnecessary work
• Measure before/after timings so any improvement is concrete and verifiable

Acceptance criteria:

• deno task check-all becomes meaningfully faster
• Full type-check coverage in the pre-commit hook is preserved
• The main sources of the slowdown are understood and documented
• The change does not reduce the reliability of local validation