Update CCIP skill guidance

[thodges-gh]

Updates the Chainlink CCIP skill guidance to clarify non-custodial workflows across CLI, SDK, MCP, CCT, monitoring, and non-EVM references. The skill now consistently prepares user-run commands, unsigned transaction data, and implementation guidance while keeping signing and broadcasting in the user’s own wallet-controlled environment.

[github-actions]

👋 thodges-gh, thanks for creating this pull request!

To help reviewers, please consider creating future PRs as drafts first. This allows you to self-review and make any final changes before notifying the team.

Once you're ready, you can mark it as "Ready for review" to request feedback. Thanks!

[thodges-gh]

Validation run:

• Current-skill smoke eval: 12/12 passing.
• Before/after regression pass against origin/main: HEAD won 8/10 affected cases, tied 2/10, with 0 baseline wins.

The new guardrails still preserve the useful safe workflow surface: discovery, status lookup, fee estimation, route/token checks, command templates, unsigned transaction construction, and code generation. The main behavior change is correctly shifting side-effecting work from “agent executes after approval” to “agent prepares user-run artifacts.”

The only meaningful follow-up I found is eval drift: several rubrics/cases still encode the old approval and second-confirmation model, so those should be updated to match the new non-custodial boundary.