Do not apply no-op schema updates

[rhuffy]

If a client submits a schema update via system_update_column_family, changes that are actually no-ops will still be announced to the cluster.

This PR skips this expensive schema update process if there is no difference in CFMetaData.

[wi11dey]

is there a risk here of skipping the update and the cluster permanently not achieving schema agreement on different uuids ? or has the uuid already been updated by this point

[andybradshaw]

is there a risk here of skipping the update and the cluster permanently not achieving schema agreement on different uuids ? or has the uuid already been updated by this point

I don't think this would affect schema agreement, as the schema doesn't actually change. If it did, this wouldn't be safe to skip. The intent of this is to prevent us from even sending the gossip messages of the no-op, as the receiving node would not actually make any writes to the CF metadata, since the mutation in this message has no modifications. I was also concerned about this, but after reading through how these messages get applied, I think this should be safe.

AFAICT, it might be safe to never announce any mutations with empty modifications, although I'm not sure how often that happens (or how expensive the commitlog activity is when receiving them). But that's more risk than we need to take to resolve the immediate issue.

[wi11dey]

ah ack. i think would be good to test on an rc and try upgrading/downgrading the old atlas to double check