CBG-3026: Use DatabaseInitManager in all GSI cases

[bbrks]

CBG-3026

Rework database initialization code to reuse DatabaseInitManager in all GSI cases (including persistent config) to centralize and de-duplicate index creation code.\

Pre-req for CBG-2838 to rearrange per-collection/system:indexes iteration logic to minimise collection overhead when building/waiting for indexes.

Depends on

• CBG3026 prereq: More low-risk cleanup/refactoring #7120

Integration Tests

• GSI=true,xattrs=true https://jenkins.sgwdev.com/job/SyncGateway-Integration/2748/
  • Known flake

[torcolvin]

There are lots of comments because I was trying to resolve the complexity. I think that there's only one bug that I can spot which is closing the connection on ClusterN1QLStore if CouchbaseCluster.GetClusterN1QLStore is called when CouchbaseCluster.bucketConnectionMode == CachedClusterConnections.

I gave a lot of commentary on parts of this code I find confusing that might be worth fixing or not. I am OK if you decide not to implement any of the other ideas, but I think it's worth consdering adding code level comments for why they aren't implemented, because I will ask them next time I read this code. This could be a problem because I'm very familiar with this code that I'm asking these questions, and someone not familiar would not ask these questions.

[torcolvin]

I like the approach but I believe there are now bugs in the bootstrap connection.

Also, the thing that needs to be fixed from the original PR is whether to make ClusterOnlyN1QLStore.Close a no-op. That depends on whether CouchbaseCluster.GetClusterN1QLStore is called from a CouchbaseCluster that has CouchbaseCluster.bucketConnectionMode == PerUseClusterConnections (It should close, otherwise ClusterOnlyN1QLStore.Close needs to be a no-op