[ESSREDUCE] refactor: WIP

[jokasimr]

Copied from #654

[SimonHeybrock]

With a new look, I think a remember our discussion from before summer (on the original PR), as well as some new comments:

1. Concerns about having model in ParameterSpec. My idea was that WorkflowSpec should be independent of workflow implementation. Having model kind of prescribes a sciline implementation.
2. use_workflow_default is similar (linked to what we have in our sciline-based workflows). I think our conclusion was that we should never use workflow defaults.
3. I think title and description should be mandatory.
4. Is ParameterSpec necessary, or would it be cleaner to just use pydantic? We may alternatively consider using a Protocol that could be fulfilled by Pydantic models.
5. WorkflowSpec: typical_outputs and other rather bespoke fields should likely be kept out of a generic spec, at least in v0. This feature was what we used in the original UI generation code, but should maybe be kept in a subclass for now (not part of this PR).
6. Is WorkflowRegistry necessary?
7. Why does WorkflowSpec hold a workflow factory? The intent of WorkflowSpec is to be generic metadata for a workflow, this seems to much tied to what is implemented concretely (based on old implementation)?

I think it is easier to have a fresh suggestion instead of editing all this into this one. I will open a PR for discussion.

[SimonHeybrock]

As announced above, I have opened #690 as a fresh take — pure-interface spec (no factory, no sciline keys), one pydantic params model per workflow, plus an ADR with the rationale. Thanks for the groundwork here; much of the analysis carried over. I propose we close this PR and continue the discussion on #690.