participate_in_coverage and restarted nodes

[martinsumner]

Scenario:

• a node is in the cluster with participate_in_coverage = true (the default);
• the node fails
• before recovering the node, it is set to participate_in_coverage = false;
• the node is restarted - and other nodes see it as an "up" node, before the ring change of that nodes coverage status is gossipped;
• the participate_in_coverage status is ignored immediately following stratup.

This is assumed to have occurred in one of the of the failure and recovery tests in Riak 3.4. See the "blip" in received vnode_index_reads on the "yellow" node when it is first restarted with coverage disabled: