Discovery: Decide on read strategy with > 1 replication for gitserver

Created by: ryanslade

When reading, how do we decide which replica to use. Possible strategies:

• Pick one at random
• Make it sticky based on user id (or anonymous id)
• Investigate how we fall back when chosen replica is down
• Can we introduce metadata in our shards column so that we track revisions per instance. Can we use a “high watermark” concept to ensure that internally we stay consistent.

Some inspiration: https://www.atlassian.com/engineering/solving-distributed-consistency-with-git

(Original JIRA: https://sourcegraph.atlassian.net/browse/COREAPP-179)