[FEAT]: Batched PDF Field Extraction (main_loop_batched)

[Harshitha-arch]

Currently, FireForm processes each PDF field sequentially with independent LLM calls, which increases latency for multi-field forms.

Rationale:
Batched processing can improve performance and scalability by sending all fields in a single structured prompt and mapping JSON output to PDF fields.

Proposed Solution:

• Add main_loop_batched() in src/llm.py.
• Use build_batch_prompt() to create one request for all fields.
• Parse JSON output and map it to PDF fields.
• Keep main_loop() unchanged for backward compatibility.
• Add minimal test in tests/test_forms.py.

Acceptance Criteria:

• Latency for multi-field PDFs decreases.
• JSON output matches schema.
• Works in Docker container.

Additional Context:
This aligns with my GSoC proposal focused on improving FireForm's pipeline efficiency.

##Fixes 399