feat: criterion micro-benchmarks for hot path

[kacy]

summary

• adds criterion benchmark suites for the two performance-critical crates on the hot path
• benchmarks keyspace get/set at multiple value sizes (64B, 256B, 1KB, 16KB) to measure raw data structure performance
• benchmarks engine routing (mpsc + oneshot channel round-trip) to measure async overhead
• benchmarks RESP3 parsing, serialization, and command dispatch to measure protocol layer cost
• adds bench, bench-core, bench-protocol Makefile targets

tested

• cargo bench -p emberkv-core -- --test — all 14 benchmark cases compile and pass
• cargo bench -p ember-protocol -- --test — all 9 benchmark cases compile and pass
• cargo clippy --workspace -- -D warnings — clean
• ran cargo bench -p emberkv-core -- keyspace_get/64B to verify real output (~45ns per GET on 64B value)

design considerations

• keyspace benchmarks operate directly on Keyspace::new() without async or channels, isolating raw data structure performance from routing overhead
• engine benchmarks use b.to_async(&rt) with a multi-thread tokio runtime to measure the real mpsc/oneshot channel cost
• protocol benchmarks test both raw parsing (byte slice → Frame) and command dispatch (Frame → Command) separately
• value sizes chosen to cover typical cache workloads: 64B (session tokens), 256B (small JSON), 1KB (document fragments), 16KB (larger payloads)