Add fast tokenizer

Generated with:

```python
from transformers import T5Tokenizer
from transformers.convert_slow_tokenizer import convert_slow_tokenizer

tokenizer = T5Tokenizer('vocabulary/spm.model', legacy=False)

fast_tokenizer = convert_slow_tokenizer(tokenizer)
fast_tokenizer.save("madlad400-3b-mt/tokenizer.json")
```

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:15aa9b3cbcf2ec220c6044f898e27fd302067a336df5369ec55a8f385f849d3b
+size 8236606