docs: explain dual-track document memory

[EricNGOntos]

Summary

• position Knowhere as a document parsing and retrieval system for agent memory
• explain Vision and Text tracks converging into one memory schema
• add a matching dual-track architecture graphic
• update model, format, and dependency guidance

Validation

• make check