feat: add gateway protocol tests and criterion benchmarks

[jbold]

Summary

• add protocol-level tests for chat.send validation and stream wiring
• add gateway integration tests covering auth required, auth rejected, and loopback no-auth flows
• add Criterion benchmark scaffolding for router resolution and sandbox cold-start calls
• switch benchmark setup to real criterion crate and register bench targets in Cargo
• sync protocol/docs/task checklist updates to reflect implemented behavior

Validation

• cargo fmt --check
• cargo clippy --all-targets -- -D warnings
• cargo test -q
• cargo bench --no-run
• cargo test --test gateway_test -- --nocapture
• cargo bench --bench router_bench -- --noplot --quick
• cargo bench --bench sandbox_bench -- --noplot --quick

Notes

• left examples/echo-plugin/Cargo.lock untracked (not part of this PR)