Update ollama_url assignment to use settings

[y0umu]

Hi developer.

The problem:
Setting the OLLAMA_URL other than the http://localhost:11434 value in .env file, or setting up DEEPSEEK_OCR_OLLAMA_URL in environment variables does not come into effect.

Example to reproduce:

export DEEPSEEK_OCR_OLLAMA_URL=http://192.168.5.8:11434
deepseek-ocr info

will get

deepseek-ocr v0.2.3
               System Information
┏━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━┓
┃ Component          ┃ Status                   ┃
┡━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━┩
│ Python             │ 3.12                     │
│ Ollama URL         │ http://192.168.5.8:11434 │
│ Ollama Running     │ No                       │
│ deepseek-ocr Model │ Not found                │
└────────────────────┴──────────────────────────┘
         Current Settings
┏━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━┓
┃ Setting          ┃ Value        ┃
┡━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━┩
│ Model            │ deepseek-ocr │
│ Output Directory │ output       │
│ Extract Images   │ False        │
└──────────────────┴──────────────┘

Supported Formats:
Images: JPG, PNG, WEBP, GIF, BMP, TIFF
Documents: PDF

⚠ Ollama is not running. Start with: ollama serve

I examined the source with the help of copilot, and notes that ollama_url argument of ModelManager.__init__ does not seem to be filled properly when ModelManager is instantialized. So the solution could be:

1. fix every instance of ModelManager
2. fix the init funtion

This pull request choose the second way. This PR fixes issues like "OLLAMA_URL" setting not working.