Fix metadata prop

[YingboMa]

-(-x) will simplify back to x and setting metadata on the simplified result is illegal.

Fixes SciML/ModelingToolkit.jl#2855

[github-actions]

Benchmark Results

	master	e16109b...	master/e16109b33608a7...
overhead/acrule/a+2	0.731 ± 0.019 μs	0.764 ± 0.022 μs	0.958
overhead/acrule/a+2+b	0.709 ± 0.02 μs	0.737 ± 0.019 μs	0.961
overhead/acrule/a+b	0.258 ± 0.0098 μs	0.265 ± 0.0088 μs	0.977
overhead/acrule/noop:Int	25.4 ± 0.93 ns	25.5 ± 0.93 ns	0.993
overhead/acrule/noop:Sym	0.0367 ± 0.0057 μs	0.0364 ± 0.0049 μs	1.01
overhead/rule/noop:Int	0.0439 ± 0.00078 μs	0.0447 ± 0.0013 μs	0.982
overhead/rule/noop:Sym	0.0562 ± 0.003 μs	0.0562 ± 0.0031 μs	0.999
overhead/rule/noop:Term	0.0563 ± 0.0027 μs	0.0589 ± 0.0032 μs	0.956
overhead/ruleset/noop:Int	0.132 ± 0.0026 μs	0.133 ± 0.0029 μs	0.991
overhead/ruleset/noop:Sym	0.158 ± 0.0061 μs	0.163 ± 0.0068 μs	0.97
overhead/ruleset/noop:Term	3.72 ± 0.14 μs	3.61 ± 0.14 μs	1.03
overhead/simplify/noop:Int	0.156 ± 0.033 μs	0.152 ± 0.0034 μs	1.03
overhead/simplify/noop:Sym	0.154 ± 0.0034 μs	0.165 ± 0.0026 μs	0.932
overhead/simplify/noop:Term	0.0381 ± 0.0021 ms	0.0371 ± 0.0016 ms	1.03
overhead/simplify/randterm (+, *):serial	0.0906 ± 0.0012 s	0.0898 ± 0.0015 s	1.01
overhead/simplify/randterm (+, *):thread	0.0533 ± 0.033 s	0.0522 ± 0.033 s	1.02
overhead/simplify/randterm (/, *):serial	0.227 ± 0.0078 ms	0.216 ± 0.0062 ms	1.05
overhead/simplify/randterm (/, *):thread	0.262 ± 0.0083 ms	0.254 ± 0.008 ms	1.03
overhead/substitute/a	0.0582 ± 0.0014 ms	0.0553 ± 0.0016 ms	1.05
overhead/substitute/a,b	0.0519 ± 0.0015 ms	0.0505 ± 0.0016 ms	1.03
overhead/substitute/a,b,c	18.6 ± 0.73 μs	18.1 ± 0.77 μs	1.03
polyform/easy_iszero	0.0342 ± 0.0017 ms	0.0321 ± 0.0017 ms	1.07
polyform/isone	2.79 ± 0.009 ns	3.1 ± 0.01 ns	0.903
polyform/iszero	1.26 ± 0.042 ms	1.19 ± 0.032 ms	1.06
polyform/simplify_fractions	1.87 ± 0.059 ms	1.83 ± 0.049 ms	1.02
time_to_load	4.69 ± 0.035 s	4.7 ± 0.029 s	0.999

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).

[YingboMa]

Tests failures seem unrelated.

[ChrisRackauckas]

Is SciML/ModelingToolkit.jl#2855 related?

[YingboMa]

It should if fix it as well.

julia> buggy(-1) # [-B₊x(t) ~ 0], as expected
1-element Vector{Equation}:
 -B₊x(t) ~ 0

julia> buggy(0)  # [0 ~ 0], as expected
1-element Vector{Equation}:
 0 ~ 0

julia> buggy(1)  # [x(t) ~ 0], but expected [B₊x(t) ~ 0]
1-element Vector{Equation}:
 B₊x(t) ~ 0

julia> buggy(2)  # [2B₊x(t) ~ 0], as expected
1-element Vector{Equation}:
 2B₊x(t) ~ 0

[ChrisRackauckas]

Well that explains that. Perfect.