feat: TensorRT-LLM Benchmark Command Generator (M113)

[hlin99]

Summary

TensorRT-LLM Benchmark Command Generator — generate ready-to-run TRT-LLM engine build, server launch, and benchmark commands for P:D ratio exploration.

Mirrors the existing vLLM and SGLang command generators.

Changes

• TRTLLMCommandGenerator class in trtllm_commands.py
• TRTLLMCommandConfig, TRTLLMServerCommand, TRTLLMBenchmarkCommand, TRTLLMCommandSet Pydantic models
• TRT-LLM specific: trtllm-build engine build + tensorrt_llm.serve server + trtllm-bench benchmark
• Options: max_batch_size, kv_cache_free_gpu_mem_fraction, pp_size, dtype, engine_dir
• Shell script output with engine build → server → benchmark → cleanup lifecycle
• CLI trtllm-commands subcommand with table + JSON + --output-script
• Programmatic generate_trtllm_commands() API
• 29 new tests

Closes #249

[hlin99-Review-Bot]

✅ Approved by hlin99-Review-Bot

Idea Value: Good — completes the TRT-LLM toolchain (importer M112 + command generator M113), consistent with vLLM and SGLang patterns.

Code Quality:

• Clean Pydantic models (TRTLLMCommandConfig, TRTLLMServerCommand, TRTLLMBenchmarkCommand, TRTLLMCommandSet)
• Generator correctly produces engine build → server → benchmark → cleanup lifecycle
• CLI registration follows established pattern
• 29 tests, all passing
• docs/iterations/current.md and ROADMAP.md updated
• CI all green (lint + tests 3.10/3.11/3.12)

LGTM 🚀

[hlin99-Review-BotX]

✅ Approved by hlin99-Review-BotX

Idea Value: Good — M113 completes the TRT-LLM command generator, consistent with the vLLM (M109) and SGLang (M111) patterns. Aligns with roadmap.

Code Quality:

• Clean Pydantic models (TRTLLMCommandConfig, TRTLLMServerCommand, TRTLLMBenchmarkCommand, TRTLLMCommandSet)
• Generator correctly produces engine build → server → benchmark → cleanup lifecycle
• CLI registration follows established pattern (register_trtllm_commands, _cmd_trtllm_commands)
• Shell script generation with proper set -euo pipefail and server PID management
• __init__.py exports and __all__ updated correctly
• ROADMAP.md and docs/iterations/current.md updated
• 29 tests, CI all green (lint + tests 3.10/3.11/3.12)

LGTM 🚀