Add n_gpu_layers to llama.cpp

[garrettsutula]

Add n-gpu-layers param to Llama.cpp model & embedding

Adds a parameter n_gpu_layers to Llama.cpp model and embedding implementation to make it possible to load & run w/ GPU. Refer to this Llama.cpp PR for more info: ggerganov/llama.cpp#1412

Who can review?

Community members can review the PR once tests pass. Tag maintainers/contributors who might be interested:

@hwchase17 @agola11

[dev2049]

thanks @garrettsutula! seems to be a duplicate of #4739 so will close if that's ok