Retrieve all care routines for a specific skin analysis batch, ordered by step number.

Also the recovery read for a lost POST /skin/care-routine: generation keeps running server-side after a gateway kills the POST response, so a client that saw the POST fail should poll this endpoint (or wait for the complete frame on the routine_progress_realtime channel advertised by POST /skin/analyze) and read status to tell "still generating" (pending) apart from "generation failed" (failed) and "ready" (completed). On the final fetch after completed, pass include_insight=1 to also receive the oyster_insight block the POST response would have carried — omit it on plain polls, it triggers an AI call.