[aggregator] Avoid large copies in entry rollup comparisons

[vdarulis]

What this PR does / why we need it:

The heavy copy takes up approx 30% of the time in both profiles and benchmarks, which is totally unnecessary.

name                  old time/op    new time/op    delta
AggregationValues-12     108ns ± 3%      76ns ± 0%  -29.35%  (p=0.008 n=5+5)

name                  old alloc/op   new alloc/op   delta
AggregationValues-12     0.00B          0.00B          ~     (all equal)

name                  old allocs/op  new allocs/op  delta
AggregationValues-12      0.00           0.00          ~     (all equal)

Special notes for your reviewer:

Does this PR introduce a user-facing and/or backwards incompatible change?:

Does this PR require updating code package or user-facing documentation?:

[robskillington]

Will this cause any new heap allocations due to escape analysis not able to see if ptr is held onto from a function called down the stack? go build -gcflags="-m" should be able to show or not I believe.

[vdarulis]

Will this cause any new heap allocations due to escape analysis not able to see if ptr is held onto from a function called down the stack? go build -gcflags="-m" should be able to show or not I believe.

@robskillington it did not, but in the end went for a different approach, as the root cause of all these copies is Go inlining, or rather lack of it - dropping this in favor of #3195, which does not change semantics and is actually even faster.