List Conversations

Return an agent's conversations with their form data and recording availability. Conversations are discovered by walking the agent's local conversation directories. There is no S3 "list" endpoint — a client's bucket only supports fetching one file by exact name — so an S3-backed conversation still gets an empty, content-free marker directory here (created at call start, see ``run_bot``) purely so it shows up in this walk at all. A row is local when its marker directory actually has ``data.json`` in it; otherwise it's read through the agent's workspace S3 config instead, resolved once per agent, not once per row. Each row reads only ``data.json`` and ``metadata.json`` — the two files the summary row actually needs. ``transcript.json``, ``metrics.json``, ``events.json``, ``recording.wav`` and ``conversation.log`` are not fetched or checked for every conversation here; those are only read when a specific conversation is opened, via the dedicated routes. Passing ``include_transcript=true`` opts back into the transcript read. **Two response shapes.** With no ``limit`` and no ``cursor`` this returns a bare JSON array of every conversation, newest first — the shape the app has always returned, and what the UI reads. Sending either ``limit`` or ``cursor`` instead returns an object:: {"data": [ ...same rows... ], "next_cursor": 20} ``next_cursor`` is ``None`` on the last page; pass it back as ``cursor`` to get the next one. The two shapes exist so paging can be added without changing a single line of the existing frontend. Every other query parameter is optional and defaults to the old behavior: no filtering, and no transcript in the rows.