feat: Support for LVM with openai compatible API with local image path

[hiro-v]

Fixes for #303
If it works, can support menloresearch/jan#294

• Support for remote image URL (contains http)
• Support for local image path (by default)
• Support for base64 image (contains data:base64)

Steps to test

1. Download models at https://huggingface.co/mys/ggml_llava-v1.5-7b/tree/main
2. Load LVM models

curl --location 'http://127.0.0.1:3928/inferences/llamacpp/loadmodel' \
--header 'Content-Type: application/json' \
--data '{
   "llama_model_path": "/Users/hiro/Downloads/ggml-model-q4_k.gguf",
   "mmproj": "/Users/hiro/Downloads/mmproj-model-f16.gguf",
   "ctx_len": 2048,
   "ngl": 100,
   "cont_batching": false,
   "embedding": false,
   "system_prompt": "",
   "user_prompt": "\n### Instruction:\n",
   "ai_prompt": "\n### Response:\n"
 }'

3. Infer with chat/completion. The image_url.url can be either base64 or local path. The remote model is not supported yet.

curl http://127.0.0.1:3928/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -d '{
    "model": "gpt-4-vision-preview",
    "messages": [
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "What’s in this image?"
          },
          {
            "type": "image_url",
            "image_url": {
              "url": "<PAYLOAD>"
            }
          }
        ]
      }
    ],
    "max_tokens": 300
  }'

[tikikun]

LGTM