Improve performance for const field

[Tortar]

This improves the performance for const fields while having the performance for mutable fields nearly equal to before.

Incredibly enough this beats base mutable structs by a huge amount!!

julia> using ProtoStructs

julia> @proto mutable struct A
           const x::Int
           const y::Int
       end

julia> mutable struct B
           const x::Int
           const y::Int
       end

julia> v_a = [A(x, x) for x in 1:1000];

julia> v_b = [B(x, x) for x in 1:1000];

julia> using BenchmarkTools

julia> @benchmark sum(x.y for x in $v_a)
BenchmarkTools.Trial: 10000 samples with 700 evaluations.
 Range (min … max):  177.974 ns … 397.629 ns  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     183.216 ns               ┊ GC (median):    0.00%
 Time  (mean ± σ):   186.304 ns ±  11.263 ns  ┊ GC (mean ± σ):  0.00% ± 0.00%

    ▃▆██▇▆▄▂▁ ▁▃▃▃▂▂▂▁▁                                         ▂
  ▂█████████████████████████▇▇▆▆▅▆▆▇▆▄▄▆▆▄▅▄▆▅▅▄▅▆▅▆▆▅▆▆▇▆▇▆▅▂▅ █
  178 ns        Histogram: log(frequency) by time        233 ns <

 Memory estimate: 0 bytes, allocs estimate: 0.

julia> @benchmark sum(x.y for x in $v_b)
BenchmarkTools.Trial: 10000 samples with 57 evaluations.
 Range (min … max):  871.544 ns …   4.771 μs  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     872.877 ns               ┊ GC (median):    0.00%
 Time  (mean ± σ):   889.405 ns ± 114.206 ns  ┊ GC (mean ± σ):  0.00% ± 0.00%

  █                                                             ▁
  █▄▄▃▃▁▁▄▄▃▁▄▄▆▇▇▆▅▅▆▆▆▆▇▄▅▅▅▅▅▄▅▃▅▅▅▅▅▅▄▃▄▄▄▃▄▄▁▁▅▁▃▄▄▄▃▃▄▄▃▄ █
  872 ns        Histogram: log(frequency) by time       1.44 μs <

 Memory estimate: 0 bytes, allocs estimate: 0.

Almost looks like something to say to some Julia core developers, but I assume there are probably good reasons why the perf difference is that big

[codecov]

Codecov Report

All modified lines are covered by tests ✅

Comparison is base (efd4ee4) 97.41% compared to head (6d7cc4c) 97.50%.
Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #27      +/-   ##
==========================================
+ Coverage   97.41%   97.50%   +0.08%     
==========================================
  Files           2        2              
  Lines         116      120       +4     
==========================================
+ Hits          113      117       +4     
  Misses          3        3              

Files	Coverage Δ
src/ProtoStruct.jl	97.47% <100.00%> (+0.08%)	⬆️

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Tortar]

Actually it seems like it works that well only when all fields are constants, but still maybe it is good to have also this opt

[Tortar]

Actually (2) with more real benchmarks I see almost no improvement, and also I still see the proto mutable version lagging behind the normal version by a lot (10x)