Native-speed vLLM transformers modeling backend

The transformers library has become a reference modeling library for machine learning that supports hundreds of architectures through consistent APIs. A modeling backend integration allows model authors to run transformers models inside vLLM without custom porting, combining transformers code with vLLM's optimized inference techniques like continuous batching and custom attention kernels. This integration now applies inference-specific layer fusions at runtime to match the speed of hand-written custom implementations for compatible architectures. Model authors can leverage this automatically by using a single flag, eliminating the need to write separate optimized code for inference deployment.