Robust Target Speaker Diarization and Separation via Augmented Speaker Embedding Sampling