Canonicalization of valid-ish tokens

[jsantell]

There are/can be scenarios where there are UCAN tokens that are almost valid (validish), and can be reasonably interpreted as such. Canonicalization of these tokens (x = serialize(deserialize(x))) is at odds with encoding valid tokens and being lenient in what we parse.

Examples of validish tokens:

• An unexpected field, either irrelevant or from a different implementation or spec/version flux
• An empty value for an optional field "prf": [] (which "MUST be omitted")
• UCAN with missing exp, re-encoded as "exp": null (feat: Allow nullable expiry per 0.9.0 spec. Fixes #23 #95 currently will trigger this scenario if given a validish token without extra handling)
• Capabilities/proofs reordered to a canonicalized order (not in spec currently, but add a note about prf sorting canonicalization#2 and JSON Canonicalization ordering could be in play)

Some options forward:

1. Strictly reject validish tokens.
2. Allow validish tokens. Retain canonicalization and encode validish tokens.
3. Allow validish tokens. Encode as valid tokens, break canonicalization.