Interactive optimizer agent

[jessica-moylan]

Summary

Allows the user to be more involved within optimization loop having the ability to manually approve/reject suggested points as well as suggesting custom points throughout the process.

Interactive Optimization Workflow through optimize_iteractively()

1. Number of optimization iterations:

2. Number of points to suggest per iteration:

3. Manually approve suggestions?

   • No — Automated suggestions
     • Run optimize_step()
     • Proceed to Step 4 (post-iteration options)
   • Yes — Manual approval mode
     • Approval cadence
       • Specify number of steps before next approval (x)
     • Per every x iterations
       • Suggest a point
       • Prompt: > “Do you approve this point?”
       • Yes → evaluate point
       • No → discard point and suggest a new one
     • After all suggested points:
       • Proceed to Step 4

4. Next action
   Prompt the user: > "What would you like to do?"

   • c — Continue optimization (no manual suggestions)
     • Return to Step 3
   • s — Suggest points manually
     • Enter DOF values
     • Ingest values into the model
     • Return to Step 3
   • q — Quit optimization

Closes #132

[thopkins32]

Looks great so far. I like the proposed flow of user input.

I know we talked about making optimize the main plan to use and branching off to other plans for interactive vs. automatic. However, I think that this won't work well. We can always refactor later if we feel they are similar enough.

Here are a few design changes:

1. Keep the original optimize intact with optimize_step the exact same as before this PR. We should only add net new functionality for now (no lines of code deleted, only added).
2. Let's do an optimize_interactive plan at a top level that removes the iterations and n_points parameter, since the user can provide those inputs interactively.
3. Remove the pytest and IPython logic in plans.py, these are not dependencies of Blop. I can help you figure out your issue with this.
4. Users must provide only DOF (aka parameter) values, never objective values. Objective values are derived from the acquired data after.

There is a remaining difficulty of getting the DOF names from the optimizer which would make the user input nice. Let's assume that we can't for now and simply have the user provide the full dictionary themselves. In a future PR we can work to improve this.