Stream query responses from boltdb index client

[sandeepsukhani]

What this PR does / why we need it:
Previously, we sent just one row per batch in the index query response callback from the boltdb index client.
This, in turn, makes us send just one row at a time per grpc response from the index gateway server.

This PR improves it by streaming the response from the boltdb index client, which would, in turn, lets us send back a larger batch in grpc response from the index gateway server.

Here is the benchmark comparing the performance difference:

benchmark                                     old ns/op      new ns/op      delta
Benchmark_QueriesMatchingLargeNumOfRows-8     2027105841     1298895890     -35.92%

benchmark                                     old allocs     new allocs     delta
Benchmark_QueriesMatchingLargeNumOfRows-8     29321207       10141758       -65.41%

benchmark                                     old bytes      new bytes     delta
Benchmark_QueriesMatchingLargeNumOfRows-8     1032313592     507791808     -50.81%

This PR also adds the benchmarks to the repository.

Checklist

• Tests updated

[cyriltovena]

LGTM

[kavirajk]

LGTM. Suggested few minor nits

[kavirajk]

do we need testing.TB here?

[sandeepsukhani]

Yes, since we use it with both tests and benchmarks.

[kavirajk]

may be worth wrapping this error? Because from the send method(), the error can be simply "timedout sending response". which I believe lacks context when I see this error in practice.

[kavirajk]

LGTM 👍