Temporal Taskification in Streaming Continual Learning: A Source of Evaluation Instability — ThinkLLM