Zero-allocation tree evaluation with buffer

[MilesCranmer]

This PR adds the ability to evaluate expression trees with zero allocations by reusing a pre-allocated buffer.

    buffer = zeros(10, size(X, 2))
    eval_options = EvalOptions(; buffer=ArrayBuffer(buffer, Ref(0)))
    result, ok = eval_tree_array(tree, X, operators; eval_options)

The buffer size must be at least (num_intermediate_results, num_samples), where num_intermediate_results depends on the tree structure. You can usually just set this equal to the maximum number of leafs.

Added tests as well.

[github-actions]

Benchmark Results

	master	0b5e4a8...	master/0b5e4a85b09271...
eval/ComplexF32/evaluation	7.26 ± 0.51 ms	7.25 ± 0.52 ms	1
eval/ComplexF64/evaluation	10.7 ± 0.95 ms	10.8 ± 1.3 ms	0.996
eval/Float32/derivative	11.9 ± 1.3 ms	12.2 ± 1.4 ms	0.977
eval/Float32/derivative_turbo	11.9 ± 1.4 ms	12.2 ± 1.5 ms	0.97
eval/Float32/evaluation	2.73 ± 0.24 ms	2.74 ± 0.24 ms	0.998
eval/Float32/evaluation_bumper	0.609 ± 0.014 ms	0.616 ± 0.014 ms	0.989
eval/Float32/evaluation_turbo	0.581 ± 0.03 ms	0.588 ± 0.029 ms	0.988
eval/Float32/evaluation_turbo_bumper	0.608 ± 0.014 ms	0.614 ± 0.014 ms	0.991
eval/Float64/derivative	15.5 ± 0.78 ms	15.7 ± 0.9 ms	0.993
eval/Float64/derivative_turbo	15.6 ± 1 ms	15.4 ± 0.74 ms	1.02
eval/Float64/evaluation	3.21 ± 0.29 ms	3.18 ± 0.33 ms	1.01
eval/Float64/evaluation_bumper	1.27 ± 0.044 ms	1.27 ± 0.042 ms	1
eval/Float64/evaluation_turbo	1.09 ± 0.059 ms	1.06 ± 0.065 ms	1.03
eval/Float64/evaluation_turbo_bumper	1.27 ± 0.041 ms	1.27 ± 0.043 ms	1
utils/combine_operators/break_sharing	0.0386 ± 0.00054 ms	0.0384 ± 0.00042 ms	1.01
utils/convert/break_sharing	27.7 ± 2.7 μs	27 ± 2 μs	1.03
utils/convert/preserve_sharing	0.0997 ± 0.0043 ms	0.0995 ± 0.004 ms	1
utils/copy/break_sharing	28.4 ± 2.3 μs	28.7 ± 2.4 μs	0.991
utils/copy/preserve_sharing	0.0997 ± 0.004 ms	0.099 ± 0.004 ms	1.01
utils/count_constant_nodes/break_sharing	9.11 ± 0.23 μs	8.88 ± 0.23 μs	1.03
utils/count_constant_nodes/preserve_sharing	0.0861 ± 0.0036 ms	0.0857 ± 0.0038 ms	1.01
utils/count_depth/break_sharing	9.76 ± 0.21 μs	9.75 ± 0.26 μs	1
utils/count_nodes/break_sharing	8.32 ± 0.21 μs	8.32 ± 0.16 μs	0.999
utils/count_nodes/preserve_sharing	0.0852 ± 0.0034 ms	0.0851 ± 0.0033 ms	1
utils/get_set_constants!/break_sharing	0.033 ± 0.0019 ms	0.0337 ± 0.0028 ms	0.978
utils/get_set_constants!/preserve_sharing	0.177 ± 0.006 ms	0.179 ± 0.0058 ms	0.991
utils/get_set_constants_parametric	0.0443 ± 0.0026 ms	0.0462 ± 0.0022 ms	0.96
utils/has_constants/break_sharing	4.06 ± 0.13 μs	4.24 ± 0.13 μs	0.958
utils/has_operators/break_sharing	2.01 ± 0.073 μs	2.02 ± 0.081 μs	0.994
utils/hash/break_sharing	22.7 ± 0.57 μs	23.3 ± 0.94 μs	0.973
utils/hash/preserve_sharing	0.0979 ± 0.0037 ms	0.0981 ± 0.0034 ms	0.998
utils/index_constant_nodes/break_sharing	24.9 ± 0.89 μs	25.6 ± 1.4 μs	0.973
utils/index_constant_nodes/preserve_sharing	0.101 ± 0.004 ms	0.102 ± 0.0038 ms	0.988
utils/is_constant/break_sharing	4.39 ± 0.14 μs	4.94 ± 0.15 μs	0.888
utils/simplify_tree/break_sharing	0.174 ± 0.0033 ms	0.174 ± 0.0027 ms	0.997
utils/simplify_tree/preserve_sharing	0.226 ± 0.0043 ms	0.223 ± 0.0045 ms	1.01
utils/string_tree/break_sharing	0.461 ± 0.015 ms	0.466 ± 0.019 ms	0.989
utils/string_tree/preserve_sharing	0.57 ± 0.02 ms	0.571 ± 0.026 ms	0.999
time_to_load	0.233 ± 0.0021 s	0.234 ± 0.002 s	0.994