feat: GC-aware allocation profiling via JFR (#118)

[rlaope]

Summary

• New argus gcprofile <pid> command that identifies which code paths allocate the most objects
• Uses JFR ObjectAllocationInNewTLAB events to rank allocation sites by bytes
• Answers: "Which code is causing GC pressure?"

Usage

argus gcprofile 12345                 # 30s default profiling
argus gcprofile 12345 --duration=60   # 60s profiling
argus gcprofile 12345 --top=20        # show top 20 sites

Changes

• GcProfileCommand — triggers JFR recording, waits, dumps, analyzes
• AllocationProfiler — parses JFR files, aggregates by stack trace top frame
• AllocationProfilerTest — unit tests for aggregation and ranking
• i18n entries (en/ko/ja)

Output Example

╭─ GC Allocation Profile ── pid:12345 ── 30s ─────────────────╮
│                                                               │
│  Total Allocated: 4.2 GB in 30s (140 MB/s)                   │
│                                                               │
│  #  Rate       Total    Stack                                 │
│  1  890 MB/s   2.4 GB   com.app.OrderProcessor.process:142   │
│  2  340 MB/s   920 MB   com.app.cache.LRUCache.put:78        │
│  3  120 MB/s   324 MB   o.a.kafka.common.record.MemRecords   │
│                                                               │
│  → OrderProcessor.process allocates 57% of total              │
╰───────────────────────────────────────────────────────────────╯

Test plan

• AllocationProfilerTest passes
• ./gradlew :argus-cli:test passes
• Manual test with live JVM

Closes #118

[rlaope]

Merged to master in 47b75d3