Docs: Add 'Why Chunkana' section with differentiators and practical limits

[asukhodko]

Motivation

• Make the README's value proposition explicit by listing concrete differentiators that matter for RAG and LLM ingestion.
• Explain common failure modes of naive splitters with a short Problem → Solution (Before/After) list to show real-world benefits.
• Surface practical, numeric defaults and limits so users can map chunking configuration to embedding and streaming constraints.

Description

• Added a new Why Chunkana section to README.md with a "Concrete differentiators" list highlighting semantic guarantees, RAG metadata, Dify compatibility, and adaptive strategies.
• Included a "Problem → Solution" (naive splitter → Chunkana) list that calls out code fences, tables, lists, and LaTeX handling.
• Documented practical defaults and limits including max_chunk_size=4096, min_chunk_size=512, overlap_size=200, and a 35% overlap cap, plus a note about streaming APIs for large files.

Testing

• No automated tests were run because this is a documentation-only change.

---

Codex Task