Refresh MiniMax M3 B200 TRT-LLM AgentX metrics (#2771)

* config(minimaxm3): refresh B200 TRT-LLM metrics

* chore: add MiniMax M3 B200 metrics refresh changelog

* fix(minimaxm3): use TRT-LLM Prometheus endpoint

* Remove obsolete TRT-LLM KV policy

* Remove unsupported MiniMax sparse options

* Fix MiniMax M3 NGC image reference

* Restore compatible MiniMax M3 TRT-LLM image

* fix(agentx): aggregate TensorRT-LLM metrics

* fix(minimaxm3): restore B200 performance tuning

* Avoid detailed TRT metrics overhead

* Fix TensorRT-LLM package lookup

* docs: consolidate metrics refresh changelog [skip-sweep]