Extended the OpenAI reasoning model client to handle different error types from the API to report explanatory/actionable error messages when a translation request fails.
Additionally, streamed translations are now disabled by default for OpenAI as streaming a reasoning model's response apparently requires a verified account. Thanks to Neoony for the efforts in diagnosing the issue!
Streaming remains the default for other providers (where supported), as this restriction seems to be unique to OpenAI.