Automatic migration of TRTLLM runtime configuration

[joostinyi]

🚀 What

• This PR improves the breaking change rollout experience for users that have previously configured runtime configs in the trt_llm.build key.

For example the following config:

trt_llm:
  build:
    base_model: llama
    checkpoint_repository:
      repo: meta-llama/Llama-3.1-8B-Instruct
      source: HF
    max_seq_len: 8192
    enable_chunked_context: False
    kv_cache_free_gpu_mem_fraction: 0.6

will have runtime configs migrated to the proper runtime key:

trt_llm:
  build:
    base_model: llama
    checkpoint_repository:
      repo: meta-llama/Llama-3.1-8B-Instruct
      source: HF
    max_seq_len: 8192
  runtime:
    enable_chunked_context: False
    kv_cache_free_gpu_mem_fraction: 0.6

with the following CLI logging:

Found extra fields ['enable_chunked_context', 'kv_cache_free_gpu_mem_fraction'] in build configuration, attempting to migrate valid fields to runtime configuration. This migration of deprecated fields is scheduled for        
removal, please upgrade to the latest truss version and update configs according to https://docs.baseten.co/performance/engine-builder-config.                                                                                   
Setting runtime.enable_chunked_context: False                                                                                                                                                                                    
Setting runtime.kv_cache_free_gpu_mem_fraction: 0.6

💻 How

• During config initialization, we look for extra build fields and attempt to patch matching runtime configuration with a deprecation notice and callout to update the truss client version.
• We will rollout this config change to the backend with 0.9.56rc2 and follow up with a full patch release of the client package.

🔬 Testing

• Added unit tests