Add benchmarks for add and sub fee dimensions

[abi87]

Why this should be merged

Dimensions are structured as value objects meaning that Add and Sub methods have a value receiver.
This causes receiver to be copied, but the receiver itself can be reused as the returned value.
On the contrary, if we used a pointer received, we'd avoid copying the receiver but we'd have to allocate the return value.
This PR introduces a couple of small benchmarks to understand what's the fastest alternative.
There seems not to be a significant different among the two alternative

Value receiver:

Benchmark_Dimensions_Add-12    	47075594	        25.24 ns/op	       0 B/op	       0 allocs/op
Benchmark_Dimensions_Sub-12    	55160683	        21.76 ns/op	       0 B/op	       0 allocs/op

Pointer receiver:

Benchmark_Dimensions_Add-12    	46841310	        25.22 ns/op	       0 B/op	       0 allocs/op
Benchmark_Dimensions_Sub-12    	55187002	        21.95 ns/op	       0 B/op	       0 allocs/op

How this works

Just added a couple of benchmarks

How this was tested

Benchmarks

[ARR4N]

Approving with comments that I don't need to re-review if implemented.

[ARR4N]

I don't think you need to explicitly call StopTimer() at the end. It's for pausing (and later resuming) in the middle.

[abi87]

Done

[StephenButtolph]

Benchmarks report zero allocations, which I understand means the compiler is able to optimize out copies.

That is not what the allocations are reporting in the benchmarks. Allocations are reporting heap allocations. Not stack copies.

[StephenButtolph]

By passing in varargs with a slice expansion, you are not copying the dimensions. The slice is passed in directly without copying.

[StephenButtolph]

Was the purpose of this PR to benchmark whether to:

• use a value receiver or a pointer receiver?
• pass in pointers for the rhs?

If we change how the rhs is passed into the benchmark, I'd expect the pointer values to be faster

[StephenButtolph]

Definitely not going to impact performance, but might as well save ~6%.

goos: darwin
goarch: arm64
pkg: github.com/ava-labs/avalanchego/vms/components/fee
                            │   old.txt   │              new.txt               │
                            │   sec/op    │   sec/op     vs base               │
_Dimensions_Add/single-12     4.469n ± 1%   4.146n ± 3%  -7.22% (p=0.000 n=10)
_Dimensions_Add/multiple-12   28.12n ± 1%   26.57n ± 5%  -5.49% (p=0.001 n=10)
_Dimensions_Sub/single-12     4.465n ± 0%   4.130n ± 3%  -7.48% (p=0.001 n=10)
_Dimensions_Sub/multiple-12   25.47n ± 1%   23.03n ± 6%  -9.60% (p=0.000 n=10)
geomean                       10.93n        10.12n       -7.46%

                            │   old.txt    │               new.txt               │
                            │     B/op     │    B/op     vs base                 │
_Dimensions_Add/single-12     0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=10) ¹
_Dimensions_Add/multiple-12   0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=10) ¹
_Dimensions_Sub/single-12     0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=10) ¹
_Dimensions_Sub/multiple-12   0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=10) ¹
geomean                                  ²               +0.00%                ²
¹ all samples are equal
² summaries must be >0 to compute geomean

                            │   old.txt    │               new.txt               │
                            │  allocs/op   │ allocs/op   vs base                 │
_Dimensions_Add/single-12     0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=10) ¹
_Dimensions_Add/multiple-12   0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=10) ¹
_Dimensions_Sub/single-12     0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=10) ¹
_Dimensions_Sub/multiple-12   0.000 ± 0%     0.000 ± 0%       ~ (p=1.000 n=10) ¹
geomean                                  ²               +0.00%                ²
¹ all samples are equal
² summaries must be >0 to compute geomean