[WIP] Add HPU support to vLLM v1

[kzawora-intel]

Early prototype. Lots of basic stuff is completely broken.

• Implemented v1 HPU attn backend, worker, model_runner and executor
• VLLM_USE_V1=1 properly selects V1 HPU components
• V1 HPU executor loads model properly
• V1 HPU executor allocates KV cache properly
• V1 HPU model runner is constructed properly and initializes bucketing
• V1 HPU attention backend gets selected automatically
• profile_run works on dummy data
• V1 HPU model_runner prepares input tensors based on SchedulerOutputs (rather than SequenceGroupMetadata)
• V1 HPU model_runner differentiates prefill and decode sequences
• V1 HPU model_runner execute_model runs for prefill
• V1 HPU model_runner execute_model runs for decode
• V1 HPU model_runner handles mixed-batch scenarios
• V1 HPU model_runner prefill returns correct results
• V1 HPU model_runner decode returns correct results
• V1 HPU model_runner warmup works properly
• Tensor parallelism works
• Capturing and replaying HPU Graphs work
• torch.compile works

To fix the problem, we need to bind the socket to a specific interface instead of all interfaces. In this case, we can bind the socket to the loopback interface (127.0.0.1), which is commonly used for local testing and does not expose the service to external networks.

• Change the binding address from '' to '127.0.0.1' on line 127.
• This change ensures that the socket only accepts connections from the local machine, mitigating the security risk.