fix(cuda): preserve replay launch compatibility

ci: port workflow cleanup from `develop-v2.1.0` to `main` (#3027)

## Summary

- ports the CI workflow cleanup from `develop-v2.1.0` to `main`
- consolidates extension and guest-library test setup while preserving
CPU, AOT, and CUDA coverage
- updates bounded runner pools, CUDA setup and caching, documentation,
release, and benchmark workflows

## Validation

- `actionlint .github/workflows/*.yml`
- workflow YAML, benchmark JSON, and script syntax checks
- `git diff --check`

release: v2.0.0 (#3000)

docs: update deferral and verify-stark book docs (#2981)

chore: update halo2 refs again (#2977)

chore: update halo2 repo refs (#2974)

docs: v2 sweep through docs (#2968)

fix: guard EVM static verifier shape assumptions (#2940)

ci: provision lint runner volume

fix(rv64): update stale zkvm cfg guard