SegINR: Segment-wise Implicit Neural Representation for Sequence Alignment in Neural Text-to-Speech