Returns the state of the benchmark's most recent prompt embedding map computation.

Restricted to platform staff, matching the action it reports on. Deliberately narrower than the map itself, which is readable by anyone who can read the benchmark: a failure reason carries the embedding provider's raw error text, so it must not reach benchmark owners or the public.

A benchmark has exactly one map, and this reports its most recent computation whatever the state, so a failure is visible here rather than silently absent. Responds with 204 No Content when no computation has ever been triggered for this benchmark.