Implement mutable addition

[blegat]

Symbolic expressions are mutable objects and exploiting this kind of mutability is essential when doing array operations like sum of a vector of matrix multiplications.
This PR is a proof of concept starting with the sum of an array of elements. We do the same with array multiplication etc... but let's start simple
The following simple benchmark:

using MutableArithmetics, Symbolics
@syms x::Real y::Real
v = repeat([x, y], 10^6)
s = @time sum(v, init = 0)
s = @time reduce(add!!, v, init = 0)

gives:

  2.669354 seconds (59.99 M allocations: 2.060 GiB, 41.07% gc time)
1000000x + 1000000y

  0.189088 seconds (2.00 M allocations: 30.503 MiB)
1000000x + 1000000y

It's still allocating a lot, I believe because a.dict[b] is type unstable since it a Dict{Any,Any} but we still save a lot of allocations.

[github-actions]

Benchmark Results

	master	f6cfdc9...	master / f6cfdc9...
overhead/acrule/a+2	1.01 ± 0.32 μs	1.08 ± 0.32 μs	0.941 ± 0.41
overhead/acrule/a+2+b	1.16 ± 0.31 μs	1.22 ± 0.31 μs	0.951 ± 0.35
overhead/acrule/a+b	0.27 ± 0.022 μs	0.259 ± 0.02 μs	1.04 ± 0.12
overhead/acrule/noop:Int	2.47 ± 0.01 ns	1.56 ± 0.01 ns	1.58 ± 0.012
overhead/acrule/noop:Sym	0.0373 ± 0.0058 μs	27.5 ± 6.4 ns	1.35 ± 0.38
overhead/rule/noop:Int	0.0361 ± 0.024 μs	0.0359 ± 0.024 μs	1.01 ± 0.94
overhead/rule/noop:Sym	0.0335 ± 0.024 μs	0.0332 ± 0.024 μs	1.01 ± 1
overhead/rule/noop:Term	0.0336 ± 0.024 μs	0.0333 ± 0.024 μs	1.01 ± 1
overhead/ruleset/noop:Int	0.124 ± 0.0046 μs	0.12 ± 0.0044 μs	1.03 ± 0.054
overhead/ruleset/noop:Sym	0.116 ± 0.0031 μs	0.123 ± 0.007 μs	0.94 ± 0.059
overhead/ruleset/noop:Term	4.57 ± 0.22 μs	4.7 ± 0.13 μs	0.974 ± 0.053
overhead/simplify/noop:Int	0.224 ± 0.015 μs	0.226 ± 0.014 μs	0.99 ± 0.089
overhead/simplify/noop:Sym	0.222 ± 0.018 μs	0.212 ± 0.0061 μs	1.05 ± 0.091
overhead/simplify/noop:Term	0.0486 ± 0.0024 ms	0.0498 ± 0.0018 ms	0.976 ± 0.06
overhead/simplify/randterm (+, *):serial	0.119 ± 0.0036 s	0.125 ± 0.0037 s	0.951 ± 0.04
overhead/simplify/randterm (+, *):thread	0.0697 ± 0.0065 s	0.0715 ± 0.0053 s	0.975 ± 0.12
overhead/simplify/randterm (/, *):serial	0.229 ± 0.022 ms	0.236 ± 0.02 ms	0.972 ± 0.12
overhead/simplify/randterm (/, *):thread	0.261 ± 0.02 ms	0.271 ± 0.019 ms	0.962 ± 0.099
overhead/substitute/a	0.101 ± 0.011 ms	0.108 ± 0.011 ms	0.932 ± 0.14
overhead/substitute/a,b	0.0885 ± 0.01 ms	0.0928 ± 0.01 ms	0.953 ± 0.15
overhead/substitute/a,b,c	20 ± 2.5 μs	21.7 ± 2.6 μs	0.922 ± 0.16
polyform/easy_iszero	0.0407 ± 0.0014 ms	0.0423 ± 0.0012 ms	0.961 ± 0.043
polyform/isone	3.1 ± 0.01 ns	2.79 ± 0.009 ns	1.11 ± 0.0051
polyform/iszero	1.37 ± 0.03 ms	1.42 ± 0.025 ms	0.96 ± 0.027
polyform/simplify_fractions	1.85 ± 0.036 ms	1.9 ± 0.029 ms	0.975 ± 0.024
time_to_load	1.15 ± 0.018 s	1.17 ± 0.0067 s	0.979 ± 0.017

Benchmark Plots

A plot of the benchmark results have been uploaded as an artifact to the workflow run for this PR.
Go to "Actions"->"Benchmark a pull request"->[the most recent run]->"Artifacts" (at the bottom).