Blop protocols and Bluesky plan refactor

[thopkins32]

Work on this started in #177.

Summary

Bluesky plans should not be so tied to the Agent like it is now. Currently it is possible to run plans in a few ways:

RE(agent.learn(...))

or explicitly what agent.learn does but on your own

suggestions = agent.suggest(...)
uid = RE(acquire(readables, dof_dict, suggestions, ...))
# <user code>
result = ...
agent.ingest(result)

I think it makes sense to remove agent.learn and offer standard Bluesky plans for optimization that follow protocols for getting new suggestions (agent.suggest) and updating models (agent.ingest). That way, users of Blop don't necessarily have to buy-in to using our Agent interface and they can decide how to perform the optimization piece by implementing the protocols.

What this means in practice

• plans.py holds standard, optimization-focused Bluesky plans
• protocols.py contains type definitions for what the above plans will use
• Agent will implement those protocols (might be renamed to something else)
• There will likely be data access and digestion protocols