Llama-cpp-http

Fixed:
- Cropping first character from llm output.

Added:
- server: DEVICE_SHUTDOWN_TIMEOUT, timeout required for device to shutdown

Changed:
- server: after each subprocess invocation to llama.cpp wait DEVICE_SHUTDOWN_TIMEOUT seconds

Changed:
- get_app() is now sync instead of async

Fixed:
- server: removed `devices_wss` and `devices_procs` that could cause GPU memory leak

Added:
- client/langchain_llama_cpp_client.py
- client/langchain_llama_cpp_embeddings_client.py
- misc/example_client_langchain_embedding.py

Changed:
- llama.cpp instructions
- `client.py` is not `client` package
- renamed misc/example_client_call_2.py to example_client_call_react.py

Fixed:
- misc/example_client_langchain_stream.py

Removed:
- misc/example_client_stream_codellama.py