research-step: tighten workflows and schemas, add autods-to-theorizer plan template

[charliemcgrady]

Summary

• Updates to the research-step skill across all three of its locations in the repo (skills/research-step, plugins/asta-preview/skills/research-step, plugins/asta/skills/research-step).
• Tightens the plan/execute/init workflows and the JSON-schema definitions in assets/schemas.yaml used by scripts/validate-output.sh.
• Adds a new autods-to-theorizer plan template under workflows/plans/ — a concrete recipe for routing AutoDS export output through the theorizer pipeline.
• Drops the auto-resolution shortcut from `plan.md`'s Replan mode: hypothesis tasks created from `literature_review` gaps are no longer silently filled in and closed. They're left open for the user / next execute pass when the source is too thin.

Notable changes

• `SKILL.md` (plugins/* copies) — refreshed description and trigger phrases.
• `workflows/plan.md` — simpler Replan table; no more inline JSON-derivation snippets.
• `workflows/execute.md` / `init.md` — updates to keep behavior consistent with the new schema.
• `assets/schemas.yaml` + `scripts/validate-output.sh` — schema additions for the autods-to-theorizer flow.

Test plan

• `scripts/validate-output.sh` passes on a representative `metadata.research_step.output` for each task_type touched
• Skill loads cleanly in Claude Code via the marketplace
• A research-step `init` against an autods export uses the new plan template