Seq-VCR: Preventing Collapse in Intermediate Transformer Representations for Enhanced Reasoning