[BP to 7.2.1] - Share plasma shards for indexes on _default scope/collection

Currently plasma is using the following scheme for co-locating indexes on shards for on-prem model:

1. index defined on _default scope and collection - dedicated shard for each index instance
2. index defined on named scope/collection - multiple index instances share a single plasma shard

Sharing the shard among multiple index instances reduces the overhead as each shard allocates a flush buffer of 2.25MB.

Consider the following single index node:

100 indexes _default scope+collection = 200 shards (back+main index) = 450MB of memory allocation for flush buffer

vs.

100 indexes in name scope/collection = 20 shards (back+main index) = 45MB of memory allocation for flush buffer

The actual logic to put multiple instances in a shard is controlled by the following flags

"indexer.plasma.maxInstancePerShard"

"indexer.plasma.maxDiskUsagePerShard"

"indexer.plasma.minNumShard"

But the general idea is that co-locating instances in a shard reduces the flush buffer overhead. 

As users can create large number of indexes on bucket(_default scope/collection) instead of using named collections, this improvement can help reduce the overheads on low configuration nodes.