Add generic AI agent workflow

[ocots]

Summary

• Rename the reusable workflow to ai-agent.yml.
• Add a generic provider input, defaulting to albert.
• Load agents by name from Handbook/ai-agents/<agent_name>.md.
• Keep deepseek as the default model alias for the initial Albert adapter.
• Validate provider and agent names before checking out the agent instructions.
• Keep provider-specific API logic inside the shared workflow so a future Claude adapter can be added without creating a workflow per agent.

The current implementation supports the Albert provider. Unsupported providers fail explicitly until their adapter is implemented.

Related PRs

• Generic agent definition in Handbook: Add first generic AI agent definition Handbook#20
• Generic CTBase manual test workflow: Add manual generic AI agent test CTBase.jl#559

Test plan

• Validate the workflow file as YAML.
• Run git diff --check.
• Execute it through the CTBase caller after the related branches are available.

Generated with Devin

[ocots]

This PR provides the reusable middle layer of the experiment:

• It consumes the agent definition from Handbook PR #20.
• It exposes the Albert model as an input, defaulting to deepseek.
• It is invoked by CTBase PR #559.

The CTBase workflow currently points to this development branch so the complete process can be tested before merging.

[ocots]

The workflow is now provider-oriented rather than agent-oriented:

• ai-agent.yml is the single reusable runner.
• provider selects the backend; albert is the first implemented adapter.
• agent_name selects a generic file such as ai-agents/hello.md.
• A future Claude integration can add another provider adapter without adding one workflow per agent.

This consumes Handbook PR #20 and is called by CTBase PR #559.