Dashboard: support Results Preview for Flotilla (distributed) queries

[samstokes]

Problem

The Results Preview tab in the dashboard only works for Swordfish (native) queries. For Flotilla (distributed Ray) queries, the tab is currently hidden (see #6557).

Why it doesn't work today

The dashboard subscriber collects result previews via on_result_out(), which requires a MicroPartition. In the Flotilla path, results are ray.ObjectRefs — calling ctx._notify_result_out() would require a ray.get() to materialize each partition, which is expensive and defeats the purpose of lazy distributed execution.

The native runner (native_runner.py:138) calls ctx._notify_result_out(query_id, result.partition()) because results are already local MicroPartitions. The ray runner (ray_runner.py:625-628) skips this entirely.

Suggested approach

Options to explore:

1. Materialize a small preview on the driver: After yielding the first few results, do a ray.get() on a small subset (e.g. first partition, head N rows) and call _notify_result_out() with that. This keeps the cost bounded.
2. Post-hoc preview: After the query finishes, materialize a small preview from the collected result partition set and send it to the dashboard.
3. Worker-side preview: Have Flotilla workers send preview data back to the driver as a side-channel, avoiding full materialization.

Context

• _notify_result_out in daft/context.py:99 explicitly raises ValueError("Query Managers only support the Native Runner for now")