Sends a single message (returns 201) or a batch of up to 1000 messages (returns 202). Messages are enqueued for delivery. Workspaces whose sending is handled inline rather than queued accept at most 200 messages per batch and send them 10 at a time, so a full batch completes well inside the request budget; a larger batch on such a workspace is refused with 400 batch_too_large before anything is sent, and should be split. Each item may carry an idempotency_key so a retried batch can be regrouped freely: matching items replay their original id with deduplicated=true, conflicting items are rejected individually, and the response counts accepted and rejected items. Authenticating with a test API key (prefix beeptest) simulates the full send lifecycle without a provisioned number and without delivering or billing anything; reserved magic numbers select deterministic outcomes (see the Testing with test keys & magic numbers guide).