Understanding the Natural Language of DNA using Encoder-Decoder Foundation Models with Byte-level Precision