refactor: self-check hallucination context bindings

Required checks

• I searched existing issues and pull requests for related refactor proposals.
• I understand that opening this issue does not mean a refactor PR will be accepted.
• I understand that a refactor PR should not be opened unless a maintainer approves the plan and assigns the work.

Problem

Copy-pasted discussion from #2253 prior to merging

This is a gap in the Manifest definition. Without looking at the self_check_hallucination() code you can't tell which keys of context are going to be used in the action.

The answer seems to be to declare two Binding.context() variables, one for bot_message and one for _last_bot_prompt. And then change the signature of self_check_hallucination() to take these directly rather than pull them out of the context dict. Also the last_bot_prompt is created by generate_bot_message() which is a Colang-only function.

The activefence rail does this and uses a RailSurface to pull out either user_message or bot_message dependeing on the name of the rail.

Fixing this problem would be a self-contained update of the self check hallucination rail to declare proper bindings and route them through at runtime rather than just context.

Proposed direction

Rewrite the self-check hallucination to use named arguments rather than puling them out of a flat dict. Need to add helper to generate last_bot_prompt outside of Colang

Migration and compatibility notes

Rewrite atomically in a single PR

Validation plan

Validate with sample prompts which contain hallucination and don't

Risks

Low - limited to just a single rail

Relevant branch or files

No response

Are you interested in helping after maintainer approval?

I am only proposing this for maintainer consideration.