In order to optimise the connections opened by a mongos process when a nss.watch() request is received, the config server will have to persist in a new collection (called config.placementHistory) historical metadata about which shards of the cluster were hosting one or more chunks of nss at a specific point in time.

The objectives of this ticket are:

• The encoding of the schema for config.placementHistory, defined as

  {
  namespace: string, // references db or collection
  timestamp: Timestamp,
  shards: array<ShardId> // may be empty
  }
  

• The instantiation of config.placementHistory as part of the initialisation of the ShardingCatalogManager (with an index over namespace+timestamp).