Major Description
EP-Engine is removing support for the time_synchronization bucket parameter. See here: MB-21144. ns_server should correspondingly remove support for this bucket parameter - in a way that does not corrupt buckets that exists with this parameter but "converts" them to buckets that don't use last-write wins conflict resolution.
This is a breaking change to the REST API (event though it's a non-advertized API) and clients, such as this one, will have to adapt their code: https://github.com/couchbase/testrunner/blob/5f943964e3fea80a2067d4c4070129e7d498f405/pytests/upgrade/upgrade_tests.py#L394.
We need to make sure that we don't break upgrade or mixed-mode cluster scenarios any worse then they have to be broken. E.g. existing LWW buckets (to the extent they exist at all) will be converted to REVID buckets. This will mean that during upgrade the conflict resolution you get is random (depending on the node you connect to.) This is unavoidable.
In mixed mode clusters, bucket creation is possibly complicated too. Certainly bucket creation must not fail. It will be possible to create a bucket with a non-disabled time_synchronization parameter that will be LWW on the not-yet-upgraded nodes and REVID on the ones that have been upgraded, and then once the upgrade completes such a freshly created bucket will no longer be LWW.
