improve performance of /canonical/leaves endpoint

[redshiftzero]

Related to #44 and should be designed together

The most performance sensitive endpoint is /canonical/leaves. That's where we provide the latest state of the enrollment list in a format that allows client-side verification. There are a lot of improvements that could be made for performance, examples:

• Strip off canonical prefix in keys - they're all the same in the canonical substore, so are redundant
• Use a more compact representation for the hex hash digests
• There are two Merkle proofs provided (I just dumped the entire proof object from cnidarium), one from the root of the canonical substore to the root (we need this), one from a "representative key" to the root of the canonical substore - we don't need this latter proof and can elide it