[Core] Support Torch profiler in Habana Worker

[mswiniarsk]

This PR allows to profile execution on HPU through flag VLLM_TORCH_PROFILER_DIR. Similar as it is done for GPU.
The profiling can be controlled:

1. Asynchronously by posting requests to the server:
   a) to start collecting profile:
   curl -X POST http://localhost:8080/start_profile
   b) to stop collecting profile:
   curl -X POST http://localhost:8080/stop_profile
2. In script, by instructing LLM object to start and stop profiling:

from vllm import LLM, SamplingParams
llm = LLM(...)
llm.start_profile()
llm.stop_profile()