FSE can be used to compress alphabets up to 12 bits in size.
Beyond that quantity, while the algorithm itself could be made to work with larger alphabets,
it would be necessary to code a specific implementation.
It's also questionable if it's a good idea, as FSE only shines when some symbols have dominant probabilities (>10%)
which is less likely when the alphabet becomes very big.