docs: add Embedding Daily Driver guide for host-runtime authors

[dgenio]

Problem

agent-kernel is an embeddable execution/authorization runtime, not a standalone proxy. The README explains the quickstart and boundaries, but host-runtime authors need a practical daily-use guide for embedding it safely.

Goal

Add a Daily Driver guide for developers adding agent-kernel to an existing agent host or tool runner.

Suggested content

• When to embed agent-kernel vs use AgentFence at the boundary.
• Minimal lifecycle: register capability -> grant token -> invoke -> expand frame -> explain trace.
• How to structure principals, roles, token TTLs, and capability scopes.
• How to handle failures fail-closed.
• How to wire policy and audit into a host runtime.
• How to compose with contextweaver, ChainWeaver, AgentFence, and weaver-spec.
• Copy-paste instructions for AI coding agents working in this repo.
• When not to use it: no clear principal model, no reviewed policy, need external proxy only.

Acceptance criteria

• Add docs/daily-driver.md or equivalent.
• Link it from README and docs index.
• Include embedding lifecycle and host-runtime patterns.
• Include when-not-to-use guidance.
• Docs/tests pass.

[dgenio]

Closing as consolidated into the new front-door/on-ramp work. Merged PR #276 now covers the embed-vs-boundary decision, low-level lifecycle, principal/authentication limits, failure/non-claims and when-not-to-use guidance; #104 owns the remaining practical “embed this in my existing host” experience. A separate Daily Driver guide would duplicate that path before #104 stabilizes.