docs: restructure for adoption — guarantees, failure model, benchmarks

[pratyush618]

Summary

Addresses external feedback: docs explain how it works but not why it's good or how to use it safely.

What changed

• Headline sharpened — "A brokerless, Rust-powered task queue for Python. Replace Celery without Redis."
• Comparison TL;DR — one-line pitch at the top of the comparison page
• Failure model — new section in architecture.md: worker crash, parent crash, DB unavailable, network partition, duplicate execution
• Delivery guarantees guide — NEW guide/guarantees.md: at-least-once semantics, why not exactly-once, idempotent task patterns (upserts, idempotency keys, unique tasks), recovery flow diagram, comparison table vs Celery/SQS
• Competitive benchmarks — throughput/latency comparison table vs Celery+Redis with "why is taskito faster" explanation
• Nav restructured — grouped by user intent: Core → Reliability → Advanced → Extensibility → Observability → Operations

What didn't change

• No existing guide pages rewritten — they're already good
• No marketing fluff — kept the code-first, practical style
• No content removed — only additions and reordering