Text this: Benchmarking bias in embeddings of healthcare AI models: using SD-WEAT for detection and measurement across sensitive populations