Can we skip check in `decode`?

[jakobnissen]

Currently, the docstring of Alphabet says:

An Alphabet's encode and decode methods must not produce invalid data.

But it is necessary to check decode? Presumably, we can simply check encode, and then we can guarantee that the internal data representation of all encodings stored in a BioSequences must be valid at any point. This way, we can skip the decoding check for increased performance.