fix: match the queried image before skipping the create
`is_recorded_in_db` returned the truthiness of the node list rather than checking
that a returned node was the image in question. The filter is only correct if the
server honours all three fields; one that ignored an unrecognised filter field
would report every image in the workspace as a match, `record_in_db` would skip
the create, and the upload would exit 0 with the image published and no record of
it - the one silent failure the exit codes exist to prevent.
Verified against the live API that all three fields are honoured, so this is not
reachable today. Matching client-side makes the check independent of filter
semantics at no cost.
Correct the recovery advice: re-running `upload` pushes again before it records,
so it is not a retry of the record. State what is broken instead.
Replace the filter test, which passed against a query that filtered on the
workspace alone, with tests that feed a non-matching node.