config replication of certain per-node keys is broken after node is offline upgraded from 2.x to 3.0 and then added back to formerly 2.x cluster that is now upgraded to 3.0

SUBJ

Steps:

1. Given: there is 2.5.1 cluster with say 2 nodes: A, B

2. Swap A to C that runs 3.0.1

3. Swap B to D that runs 3.0.1

4. upgrade package on A from 2.5.1 to 3.0.1

5. Add A back to cluster

6. Observe that C and D don't have certain per-node config keys of A. Specifically capi_port

It causes node info for that added back node (A in this case) to miss important fields (capi_port, ssl ports) if bucket info (terse or otherwise) is requested from nodes C or D.

(TODO: verify that removing config.dat doesn't cause the problem. I.e. problem depends on existence of old config.dat with config_version being 2.5 (or 2.0, most likely too))

Workaround: when doing online upgrade to 3.x always delete package fully (including config files) before installing 3.x package.