Harden benchmark workflow input handling

[fegge]

Summary

• Validate manual benchmark workflow inputs before using them in shell or make command contexts.
• Pass benchmark action inputs through environment variables and arrays instead of direct shell interpolation.
• Harden local benchmark CLI and formatter script argument handling.

Validation

• ruff format scripts/tests: passed
• ruff check scripts/tests: passed
• python3 -m py_compile scripts/tests: passed
• git diff --check -- .github/actions/bench/action.yml .github/workflows/bench_ec2_any.yml .github/workflows/bench_ec2_reusable.yml scripts/format scripts/tests: passed
• bash -n scripts/format: passed
• shellcheck scripts/format: passed
• actionlint .github/workflows/bench_ec2_any.yml .github/workflows/bench_ec2_reusable.yml: passed
• zizmor .github/workflows/bench_ec2_any.yml .github/workflows/bench_ec2_reusable.yml .github/actions/bench/action.yml: passed with no findings
• Unsafe --cflags, --ldflags, and --cross-prefix benchmark CLI probes: rejected with exit code 2 and did not create marker files
• Not run: ./scripts/format because nixpkgs-fmt is not installed in this local environment
• Not run: ./scripts/lint because shfmt is not installed in this local environment

Fixes #1213

---

This work was completed by Trail of Bits as part of the Patch The Planet project in collaboration with OpenAI. The issue was identified primarily by the Codex coding agent, and manually reviewed before submission.

[fegge]

Reopened from an upstream repository branch with DCO-signed commits so full CI can run: #1233