Storage should reject read requests for non-existent org-bucket pairs

I was writing some tests for task run logs reading from the system bucket, and when I used a deliberately bad organization ID, against a real storage engine and query controller, I was surprised to see that the query returned successfully (albeit with an empty result set).

Reading from a non-existent org-bucket pair is roughly equivalent to reading from a non-existent DB in 1.x. The read should be rejected with a useful error message.

Start by reproducing the task-run-log read against the system bucket with a deliberately invalid organization ID, using a real storage engine and query controller. Trace the storage read path and query-controller handling to determine where a missing org-bucket pair currently becomes an empty result. Done means the read is rejected with a useful error while existing valid reads continue to work.

Written by the indexing model from the issue text.