feat: check that the CidBuilder hasher is usable

[rvagg]

Ref: #90

When using SetCidBuilder you can set up for a hasher that will not be found when a Sum() is eventually called in EncodeProtobuf(), and the error from this doesn't bubble up through RawData(), it just panics.

So this change adds an error return to SetCidBuilder() and will fail to use the builder if it's going to result in that error. If the error isn't handled then you'll get a silent default to a (presumably) working builder. If it's checked then at least you'll have an error to bubble up to the user or some place helpful.

However, the caveat here is that we are not actually testing Sum() will work, only the obvious path to the error. If we are given a custom CidBuilder that doesn't match Prefix, or if there is a custom Sum() that doesn't use MhType then this isn't going to work. But I'm not sure what that situation is.

This is a soft option for dealing with #90.

[Jorropo]

This is better, now all consumers are protected to what we found in Kubo.

[rvagg]

@Jorropo got an opinion on version bump for this? I'm not really sure how a non forward-breaking in a <0 semver situation would normally be handled. In >1 I'd probably just bump minor version for this so I'm tempted to say in <0 that it's a patch bump and let static analysis pick up the change for consumers.

[rvagg]

Needs to be a semver-minor bump because it's a method so it impacts the type's interface.

I'm also considering further testing that if the builder isn't a cid.Prefix running a Sum() on it with a simple value (empty byte array?) to test that it actually works.

[Jorropo]

@rvagg I don't fully understand where is the API changing here ?
Because it wouldn't work either way, before it would panic later, but now it's an error earlier.
AFAIT there is no way to use this code with broken hasher, before it would panic or error when getting bytes, now it's when setting a build.

I'm also considering further testing that if the builder isn't a cid.Prefix running a Sum() on it with a simple value (empty byte array?) to test that it actually works.

Yeah, this would also work in-case a custom builder is passed, I think it's fine.