Calculate and maintain collection size and row count

[efirs]

• Add a machinery for low overhead atomic increment
• Add a KV layer with statistics calculation on write APIs.
• Use the same KV layer for calculation of standalone search (so as we persist it in FDB as well)
• Maintain separate set of atomics at query_runner level to calculate inline search size statistics.

They APIs which doesn't have enough information in the request itself (replace,delete), will read the value first.
We can apply various optimization, like reading first chunk only and not unmarshalling/decompressing the payload.

[himank]

@efirs should we close this?