This PR improves the benchmark suite to better reflect real throughput limits and isolate different bottlenecks:

1. Multi-connection gRPC benchmark:

   • Adds a variant where each benchmark worker uses its own gRPC connection
   • Removes the single-connection bottleneck and allows the server to scale with available CPUs

2. PURE ECDSA signing benchmark over gRPC

   • Adds a Pure ECDSA benchmark that exercises real CPU and allocation pressure

3. Improved plots

   • Updated plots to better compare CPU scaling and GOGC sweeps
   • Catch /direct and /grpc benchmarks