MQE: don't create copies of native histogram during binary operations

[charleskorn]

What this PR does

This PR removes creating copies of native histogram during binary operations in MQE.

This dramatically improves the performance of binary operations involving native histograms, for example:

                                                                    │ original.txt  │              after.txt              │
                                                                    │    sec/op     │    sec/op     vs base               │
Query/nh_1_/_a_1,_instant_query/engine=Mimir-10                        312.2µ ±  1%   330.7µ ±  2%   +5.96% (p=0.002 n=6)
Query/nh_1_/_a_1,_range_query_with_100_steps/engine=Mimir-10           373.7µ ±  3%   374.2µ ±  2%        ~ (p=0.699 n=6)
Query/nh_1_/_a_1,_range_query_with_1000_steps/engine=Mimir-10         1051.5µ ±  2%   885.6µ ±  2%  -15.77% (p=0.002 n=6)
Query/nh_100_/_a_100,_instant_query/engine=Mimir-10                    5.440m ±  4%   5.268m ±  0%   -3.16% (p=0.002 n=6)
Query/nh_100_/_a_100,_range_query_with_100_steps/engine=Mimir-10      11.400m ±  5%   9.901m ±  1%  -13.15% (p=0.002 n=6)
Query/nh_100_/_a_100,_range_query_with_1000_steps/engine=Mimir-10      61.10m ±  5%   51.58m ±  1%  -15.58% (p=0.002 n=6)
Query/nh_2000_/_a_2000,_instant_query/engine=Mimir-10                 101.00m ±  7%   99.37m ±  5%        ~ (p=0.485 n=6)
Query/nh_2000_/_a_2000,_range_query_with_100_steps/engine=Mimir-10     204.6m ±  5%   178.1m ±  2%  -12.96% (p=0.002 n=6)
Query/nh_2000_/_a_2000,_range_query_with_1000_steps/engine=Mimir-10   1189.4m ±  1%   970.7m ±  5%  -18.39% (p=0.002 n=6)
Query/nh_1_/_2,_instant_query/engine=Mimir-10                          194.8µ ±  7%   182.9µ ±  3%   -6.08% (p=0.002 n=6)
Query/nh_1_/_2,_range_query_with_100_steps/engine=Mimir-10             269.9µ ±  6%   226.2µ ±  3%  -16.20% (p=0.002 n=6)
Query/nh_1_/_2,_range_query_with_1000_steps/engine=Mimir-10            870.9µ ±  3%   682.9µ ±  3%  -21.59% (p=0.002 n=6)
Query/nh_100_/_2,_instant_query/engine=Mimir-10                        4.700m ±  5%   4.525m ±  1%   -3.73% (p=0.002 n=6)
Query/nh_100_/_2,_range_query_with_100_steps/engine=Mimir-10          10.217m ±  1%   8.589m ±  1%  -15.93% (p=0.002 n=6)
Query/nh_100_/_2,_range_query_with_1000_steps/engine=Mimir-10          56.92m ±  6%   46.10m ±  1%  -19.02% (p=0.002 n=6)
Query/nh_2000_/_2,_instant_query/engine=Mimir-10                       85.43m ±  4%   83.10m ±  1%   -2.72% (p=0.002 n=6)
Query/nh_2000_/_2,_range_query_with_100_steps/engine=Mimir-10          179.7m ±  7%   153.1m ±  1%  -14.80% (p=0.002 n=6)
Query/nh_2000_/_2,_range_query_with_1000_steps/engine=Mimir-10        1091.5m ±  1%   820.6m ±  2%  -24.82% (p=0.002 n=6)

This has no noticeable impact on binary operations only on floats, and has no noticeable impact on peak memory consumption.

Which issue(s) this PR fixes or relates to

(none)

Checklist

• [n/a] Tests updated.
• [n/a] Documentation added.
• [n/a] CHANGELOG.md updated - the order of entries should be [CHANGE], [FEATURE], [ENHANCEMENT], [BUGFIX].
• [n/a] about-versioning.md updated with experimental features.

[charleskorn]

This is ready for a review, but will require changes to support group_left and group_right after #10119 is merged.

[jhesketh]

lgtm, but will approve once reworked with the group changes.

I haven't looked closely, but do we have good tests to that would fail if we weren't nil'ing values before returning to the pool appropriately? (Just thinking how we can better test this).

I also imagine this would not work if we every revisited trying not to copy histograms on lookbacks. It might be worth adding a comment somewhere as such.

[jhesketh]

What do you think about keeping this method as just returning the next values.

Then have a couple of extra methods to getCurrentHIndex / getCurrentFIndex ?

I feel like that would be neater and keep the methods clearer on their uses.