feat: argus benchmark — JMH-like micro-benchmark with GC/allocation profiling

[rlaope]

Priority: P3

Perspective: Performance Engineer

Why

"I want to benchmark a method and see not just throughput, but GC impact, allocation rate, and CPU profile in one report."

Design

argus benchmark <pid> com.example.Serializer.serialize --iterations=1000

Output:

╭─ Benchmark Report ──────────────────────────────────────────╮
│                                                              │
│  Method: com.example.Serializer.serialize                    │
│  Iterations: 1,000                                           │
│                                                              │
│  Throughput     12,345 ops/s                                 │
│  Avg Latency    81μs (p50: 72μs, p99: 340μs)               │
│  Allocations    2.3 MB/iteration (1,245 objects)            │
│  GC Triggered   3 times during benchmark                    │
│  GC Overhead    1.2% of benchmark time                      │
│                                                              │
│  Top Allocating Frames:                                      │
│    1. ObjectMapper.writeValueAsBytes:123  (890KB/iter)       │
│    2. ByteArrayOutputStream.grow:78      (460KB/iter)        │
╰──────────────────────────────────────────────────────────────╯

Impact

Combines JMH + allocation profiling + GC analysis in one command. No competitor offers this.