Unified probe executions endpoint. Behavior:

• Accepts a mixture of search/sort/filters and returns enriched ProbeExecutionOut rows.
• Inputs:
  • query: optional vector search query (non-persisted). When present, the endpoint uses the domain’s probe vector store to find related execution ids and returns those rows.
  • search_id: optional id of a previously saved vector search; when supplied, the endpoint returns rows for that saved search id. This enables stable “recent searches” or revisiting prior results.
  • models: repeated filter on ProbeExecution.model (e.g., models=gpt-5&models=gemini)
  • objective_ids: repeated filter on ProbeTemplate.objective_id (UUIDs)
  • sort: asc | desc (on executed_at)
• Output:
  • { search_id?: string, results: ProbeExecutionOut[] }
  • ProbeExecutionOut includes model, question_text, response_text, reasoning, citations, purpose, objective_id, etc.
• Selection logic:
  1. If search_id provided: load execution_ids for that search and return those rows with filters/sort applied server-side.
  2. Else if query provided: perform an immediate vector search (non-persisted), map results to execution_ids, and return rows.
  3. Else: return the latest execution per probe for the domain, with any filters applied (models/objective_ids).