The documentation does describe a fairly flexible parser. Perhaps it's a little too flexible on stuff like URN prefixes, but I don't think we could start enforcing a stricter class of hyphen separations without potentially breaking existing code.

Is there are reason your validator doesn't use uuid.UUID to normalize the value? That is, whatever the customer provides, why not use the result of stringifying the resulting UUID, rather than just convert to UUID to validate, then throwing it away? As long as the result is compatible with your sql connector, and logically equivalent to what the customer provided, that seems a valid solution.