[FEAT] Implement Batch LLM Extraction for 10x Performance Gain

[lokesh12344]

Is your feature request related to a problem?

The current LLM extraction logic makes sequential HTTP requests to Ollama for every single field in a PDF. On CPU-based systems, this is extremely slow (~1-2 minutes per field), making the app difficult to use for standard forms with 5+ fields.

Describe the solution you'd like

I propose a "Batch Extraction" engine that:

1. Consolidates all field requests into a single, structured prompt.
2. Utilizes Ollama's format: "json" mode for structured output.
3. Extracts all data in one single LLM call.

Describe alternatives you've considered

Keeping the sequential model, but it is too slow for production use on standard hardware.

Additional context

Preliminary testing shows that this reduces extraction time from ~10 minutes down to ~1 minute for a standard form (a 10x speedup). This also improves contextual accuracy as the LLM sees all fields at once.