Add TFLOP pipeline walkthrough documentation

[ankitaggnitt]

This commit adds a comprehensive Markdown document that provides a step-by-step walkthrough of the TFLOP pipeline, from data preprocessing to model architecture and training/evaluation. It bridges the concepts from the research paper with their concrete implementations in the Python codebase, citing relevant files and code snippets.

Key areas covered:

• Data Preprocessing: Raw inputs, HTML to OTSL conversion, text region processing (bounding box normalization, layout embedding), pointer target creation, and final batch assembly.
• Model Architecture: Overall TFLOP class structure, Image Encoder (Swin), Layout Encoder components, Logical Structure Decoder (MBART), Layout Pointer mechanism (pointer head, dot-product similarity, pointer loss, empty cell handling), and Span-aware Contrastive Supervision (loss module, positive/negative set construction, span coefficients, final contrastive loss).
• Training/Evaluation: Main training script, key arguments, optimizer/scheduler, inference process (autoregressive generation, pointer usage, HTML construction), and TEDS evaluation details (use of apted library).

[ankitaggnitt]

Let me know if its any good.