fix: remove Hash + PartialEq on ConfirmedOrder and dependent structs

[huitseeker]

What:

Remove impls of Hash, PartialEq, Eq on CertifiedOrder and dependents

Why

It's a quick and easy way to check that certificates can't be used in a way that assumes the details of their signature set are meaningful. Certificate equivocation (change in signature set without loss of overall validity) is allowed.

We might indeed need to compare CertifiedOrders in the future, in which case a comment at the point where a static_assertions will blow up offers guidance on what to do.

Why not do more?

The reason this doesn't already implement the signature-agnostic way of comparing Certificates is to avoid misuse of that Eq implementation in a context where we do not have distinct ValidCertifiedOrder and UncheckedCertifiedOrder types.
Imagine you have a list of CertifiedOrders L and an Eq function that's signature agnostic. You also have a deduplicate function on vecs that uses Eq:

• if all certificates are valid, should you call dedup? Probably yes,
• if all certificates have yet to be checked, should you call dedup? Probably not.

But there's a lot of duplication of utility functions!

Yes. Those are used in a test context where doing the dangerous thing (testing certificates byte for byte) makes sense, either because one of:

• we check correct ser/de,
• we check an authority has not led to an update of our stores as a client, or of its own stores as an authority, under what should be an idempotent op

This could be implemented with less code duplication using a test feature (to make sure the utility functions can be exported across crates but only in a test context), but would lead to fighting Rust's feature unification. The ad-hoc utility functions are actually less trouble.

Contributes to #266

[sblackshear]

Thanks for this!!