Implement a BPE tokenizer from scratch: train merge rules, then encode and decode without external libraries.

Count adjacent symbol pairs, repeatedly merge the most frequent pair, and store the ordered merge table. Encoding reapplies merges by rank; decoding concatenates token bytes, so normalization and unknown-byte handling must be explicit.

Test empty input, one-element input, duplicates, boundary indices, invalid states, and the largest allowed size; state time and space complexity.