feat: Add comprehensive benchmark suite for RPC performance testing

[achimnol]

resolves lablup/backend.ai#184

Summary

• Add a full-featured benchmark suite to measure and analyze callosum RPC performance
• Include throughput, latency, and feature overhead benchmarks with multiple scenarios
• Provide CLI interface, console/HTML reporters, and CPU/memory profiling support
• Remove unused imports from lower transport modules

Benchmark Scenarios

Throughput:

• Payload size scaling (64B to 1MB)
• Client concurrency (1 to 100 clients)
• Scheduler comparison (exit-ordered vs key-serialized)

Latency:

• Load testing with configurable target RPS
• Payload size impact analysis
• Tail latency (p99, p99.9) tracking

Feature Overhead:

• Snappy compression impact
• CURVE authentication overhead
• Combined features matrix

Usage

# Install benchmark dependencies
uv sync --extra benchmark

# Run all benchmarks
uv run python -m benchmarks

# Run specific scenario
uv run python -m benchmarks --scenario throughput

# Generate HTML report
uv run python -m benchmarks --output ./results --format html

Test plan

• Verify mypy passes: uv run mypy benchmarks/
• Run a quick benchmark test: uv run python -m benchmarks --scenario throughput
• Check HTML report generation works correctly