Improve AI agent quickstart and examples

[melonattacker]

Motivation

logira's strongest initial use case is local auditing of AI coding agents such as Claude and Codex.

The README should make this use case immediately obvious.

A new user should quickly understand:

• why they would use logira with Claude/Codex
• how to run an agent under audit
• how to inspect a run
• how to interpret detections
• what logira does and does not do

Goal

Improve documentation for local AI agent workflows.

Scope

Add or improve docs for:

• Claude usage
• Codex usage
• strong-permission / yolo-style workflows
• what logira observes
• what logira does not observe
• how to inspect a run
• how to export a report if report support exists
• observe-only design
• local-only storage

Proposed docs

README should include a short quickstart.

Optional detailed docs:

docs/ai-agents.md
docs/reports.md

Example quickstart

logira status
logira run -- claude
logira view last
logira explain last --show-related

If report support exists:

logira report last --format markdown

Non-goals

This issue should not include:

• team deployment docs
• hosted UI docs
• cloud docs
• long marketing copy

Acceptance criteria

• README clearly states the AI coding agent use case.
• docs include Claude example.
• docs include Codex example.
• docs explain local-only storage.
• docs explain observe-only behavior.
• docs explain how to inspect a run.
• docs mention that logira records OS-level behavior rather than relying on the agent's own narrative.

Notes

This is a should-have for v0.2.0.

It should stay small and practical.