Benchmark suite

What about adding benchmarks to the repo ? Here is the beginning of a benchmark suite.

I was unable to find any convenient tools for formatting and organizing reports. Whats more, after poking around in a few packages, it seems that these things tend to fall behind the rest of the code.

But, it helps me keep track of what has been done. For instance, here are the relevant benchmarks before the pending PR:

julia> println.(result["Eye"]["identities"]);
Pair{Any,Any}(("inv", "Int64", "n=1000"), Trial(960.983 ns))
Pair{Any,Any}(("inv", "Float64", "n=10"), Trial(53.983 ns))
Pair{Any,Any}(("LinearAlgebra.triu", "Int64", "n=10"), Trial(55.983 ns))
Pair{Any,Any}(("permutedims", "Float64", "n=1000"), Trial(6.776 ms))
Pair{Any,Any}(("inv", "Int64", "n=10"), Trial(52.983 ns))
Pair{Any,Any}(("LinearAlgebra.tril", "Float64", "n=10"), Trial(60.983 ns))
Pair{Any,Any}(("LinearAlgebra.tril", "Float64", "n=1000"), Trial(718.983 ns))
Pair{Any,Any}(("LinearAlgebra.triu", "Float64", "n=10"), Trial(54.983 ns))
Pair{Any,Any}(("inv", "Float64", "n=1000"), Trial(970.983 ns))
Pair{Any,Any}(("LinearAlgebra.tril", "Int64", "n=10"), Trial(55.983 ns))
Pair{Any,Any}(("permutedims", "Int64", "n=10"), Trial(1.538 μs))
Pair{Any,Any}(("permutedims", "Float64", "n=10"), Trial(1.543 μs))
Pair{Any,Any}(("permutedims", "Int64", "n=1000"), Trial(6.678 ms))
Pair{Any,Any}(("LinearAlgebra.triu", "Int64", "n=1000"), Trial(619.983 ns))
Pair{Any,Any}(("LinearAlgebra.triu", "Float64", "n=1000"), Trial(679.983 ns))
Pair{Any,Any}(("LinearAlgebra.tril", "Int64", "n=1000"), Trial(705.983 ns))

Here are the same benchmarks after the PR:

julia> println.(result["Eye"]["identities"]);
Pair{Any,Any}(("inv", "Int64", "n=1000"), Trial(0.001 ns))
Pair{Any,Any}(("inv", "Float64", "n=10"), Trial(0.001 ns))
Pair{Any,Any}(("LinearAlgebra.triu", "Int64", "n=10"), Trial(0.001 ns))
Pair{Any,Any}(("permutedims", "Float64", "n=1000"), Trial(0.001 ns))
Pair{Any,Any}(("inv", "Int64", "n=10"), Trial(0.001 ns))
Pair{Any,Any}(("LinearAlgebra.tril", "Float64", "n=10"), Trial(0.001 ns))
Pair{Any,Any}(("LinearAlgebra.tril", "Float64", "n=1000"), Trial(0.001 ns))
Pair{Any,Any}(("LinearAlgebra.triu", "Float64", "n=10"), Trial(0.001 ns))
Pair{Any,Any}(("inv", "Float64", "n=1000"), Trial(0.001 ns))
Pair{Any,Any}(("LinearAlgebra.tril", "Int64", "n=10"), Trial(0.001 ns))
Pair{Any,Any}(("permutedims", "Int64", "n=10"), Trial(0.001 ns))
Pair{Any,Any}(("permutedims", "Float64", "n=10"), Trial(0.001 ns))
Pair{Any,Any}(("permutedims", "Int64", "n=1000"), Trial(0.001 ns))
Pair{Any,Any}(("LinearAlgebra.triu", "Int64", "n=1000"), Trial(0.001 ns))
Pair{Any,Any}(("LinearAlgebra.triu", "Float64", "n=1000"), Trial(0.001 ns))
Pair{Any,Any}(("LinearAlgebra.tril", "Int64", "n=1000"), Trial(0.001 ns))

Start by reviewing the proposed benchmark/benchmark/FillArraysBenchmarks.jl file and the benchmark results shown in this issue. Determine how benchmarks should be organized and reports formatted; done means the repository has a usable benchmark suite that tracks the relevant operations and produces organized results.