add num builder gpus config

[joostinyi]

:rocket: What

• We have observed that for certain build configurations (e.g. FP8 quantization) the build time and runtime resources can differ, necessitating independent configurability of these resources.

  :computer: How

• Add a new config trt_llm.build.num_builder_gpus
• Add associated warning for unset num_builder_gpus when using FP8 quantization types

  :microscope: Testing

[linear[bot]]

BT-11273 Engine builder: FP8 quantization needs more GPU memory than model, support this