getOrPutAll

The batch form of getOrPut: looks up many prompts at once, embedding them in a single Embedder.embedAll call.

Most embedding providers price and rate-limit per request, not per token, so embedding fifty prompts in one call is far cheaper and faster than fifty calls. Each prompt is then looked up with its vector, and every miss is computed with compute and cached, exactly as getOrPut does; the returned responses line up with prompts.

This is a batch primitive, so it trades a little of getOrPut's cleverness for the batch win: concurrent-miss coalescing and the negative cache (both keyed on a single prompt) are skipped, and misses are computed one after another in order — wrap compute yourself if you want the model calls to run concurrently. Writes still honour write-behind when it is on. embedFailurePolicy applies to the batch embed the same way it does to a single one.