fix: Improve PDF parser robustness and efficiency

[Dhruv-Sharma01]

Closes #137

Changes Implemented
This PR introduces a more robust, two-stage parsing architecture. Below are the specific changes made to each file.

1. In pymupdf_rag.py (PDF-to-Markdown Conversion):

Replaced IdentifyHeaders class with AdvancedHeaderIdentifier: The new class uses a multi-heuristic scoring system (boldness, all-caps, relative font size) to detect section headers, making it more resilient to different resume styles.

Enforced Linear Reading Order: Added a sort for the text_rects list (text_rects.sort(key=lambda r: (r.y0, r.x0))) immediately after text blocks are identified. This ensures a strict top-to-bottom, left-to-right processing flow, fixing parsing errors on multi-column layouts.

2. In pdf_handler.py (Section Extraction):

Added _split_markdown_by_headers method: This new helper function provides a deterministic way to pre-process the Markdown text, splitting it into a dictionary of sections based on ## headers before any LLM calls are made.

Refactored _extract_all_sections_separately method: The original logic, which made multiple, full-document LLM calls, has been replaced. The new implementation first uses _split_markdown_by_headers to get structured data and then sends only the small, relevant text chunk for each section to the LLM for analysis. This improves efficiency and reliability.

[Mohd-Mursaleen]

You messed up the formatting use Black.

[Dhruv-Sharma01]

Check it.

[Mohd-Mursaleen]

Check it.

👍🏻