Add a stable public interface for accessing the data underlying a BenchmarkResult

[DilumAluthge]

Fixes #8

The idea here is that we add a stable public interface for getting the raw data inside a BenchmarkResult. Then users can just take that raw data and use whatever plotting package they want.

[chriselrod]

Do you want to get this in before registering?
Or maybe change the plotting library?

[DilumAluthge]

Let's get this in first. I'll do one last test locally and then I'll mark this as ready for review.

I think the default plotting library is good, no need to change it.

[DilumAluthge]

Just as an example of how to use this feature. I ran the following code:

julia> import BLASBenchmarksCPU

julia> import StatsPlots

julia> benchmark_result = BLASBenchmarksCPU.runbench(Float64; sizes = [1, 2, 5, 10, 20, 50, 100, 200, 500, 1_000])
Progress: 100%|███████████████████████████████████████████████████| Time: 0:06:46
  Size:      (1000, 1000, 1000)
  BLIS:      68.47
  Gaius:     44.62
  MKL:       69.59
  Octavian:  61.52
  OpenBLAS:  66.8
  Tullio:    51.29
Bennchmark Result of Matrix{Float64}, threaded = true
60×4 DataFrame
 Row │ Size   Library   GFLOPS        Time
     │ Int64  Symbol    Float64       Float64
─────┼────────────────────────────────────────────
   1 │     1  BLIS       0.000117592  1.7008e-5
  ⋮  │   ⋮       ⋮           ⋮             ⋮
  60 │  1000  Tullio    51.2924       0.0389921

julia> df = BLASBenchmarksCPU.benchmark_result_df(benchmark_result)
60×4 DataFrame
 Row │ Size   Library   GFLOPS        Time
     │ Int64  Symbol    Float64       Float64
─────┼────────────────────────────────────────────
   1 │     1  BLIS       0.000117592  1.7008e-5
  ⋮  │   ⋮       ⋮           ⋮             ⋮
  60 │  1000  Tullio    51.2924       0.0389921

julia> df[!, :Size] = Float64.(df[!, :Size])
60-element Vector{Float64}:
    1.0
    ⋮
 1000.0

julia> df[!, :GFLOPS] = Float64.(df[!, :GFLOPS])
60-element Vector{Float64}:
  0.00011759172154280339
  ⋮
 51.29240051024655

julia> df[!, :Time] = Float64.(df[!, :Time])
60-element Vector{Float64}:
 1.7008e-5
 ⋮
 0.038992131

julia> StatsPlots.@df df StatsPlots.plot(
           :Size,
           :GFLOPS;
           group = :Library,
           legend = :bottomright,
       )

And I received as output this plot, which was generated by StatsPlots.jl:

[DilumAluthge]

This is ready to merge.

[chriselrod]

Looks like performance could be a lot better.