Returns the benchmark's prompt embedding map.

Unpaged: the whole corpus arrives in one response because the plot needs every point at once. Tag strings are de-duplicated into a map-level tags vocabulary that each point indexes via tagIndices. Prompts without English text are not embedded, so promptCount can be lower than the benchmark's prompt total and the map is not a complete corpus listing. Responds with 204 No Content when no map has been computed yet.