Rewrite 4x4 det as product of 2x2 minors for improved performance

[snoeyink]

The SMatrix{4,4} det code (added in #260) performs 72 multiplications and 23 addition/subtractions. Rewriting as products of 2x2 minors reduces to 30 multiplications and 17 additions/subtractions. Benchmarking (below) shows performance improvement from 11ns to <5ns, commensurate with the decrease in operations.

I see that in pull request #597 @ryanelandt (with the approval of @andyferris) implicitly rewrote inv using 2x2 minors. Thus, my pull request just extends the performance improvement to det.

---

julia> using BenchmarkTools, Random, LinearAlgebra                                                        
                                                                                                          
julia> using StaticArrays                                                                                 
[ Info: Precompiling StaticArrays [90137ffa-7385-5640-81b9-e52037218182]                                  
                                                                                                          
julia> function bmold()                                                                                   
       N=4                                                                                                
       Random.seed!(1234)                                                                                 
       SA = @SMatrix rand(N,N)                                                                            
       A = Array(SA)                                                                                      
       @assert(det(SA) ≈ det(A))                                                                          
       @benchmark StaticArrays._olddet($(Size(SA)), $(SA))                                                
       end                                                                                                
bmold (generic function with 1 method)                                                                    
                                                                                                          
julia> function bm()                                                                                      
       N=4                                                                                                
       Random.seed!(1234)                                                                                 
       SA = @SMatrix rand(N,N)                                                                            
       A = Array(SA)                                                                                      
       @assert(det(SA) ≈ det(A))                                                                          
       @benchmark StaticArrays._det($(Size(SA)), $(SA))                                                   
       end                                                                                                
bm (generic function with 1 method)                                                                       

julia> bmold()
BenchmarkTools.Trial: 10000 samples with 999 evaluations.
Range (min … max): 10.975 ns … 47.887 ns ┊ GC (min … max): 0.00% … 0.00%
Time (median): 11.259 ns ┊ GC (median): 0.00%
Time (mean ± σ): 11.398 ns ± 1.592 ns ┊ GC (mean ± σ): 0.00% ± 0.00%
Memory estimate: 0 bytes, allocs estimate: 0.

julia> bm()
BenchmarkTools.Trial: 10000 samples with 1000 evaluations.
Range (min … max): 4.582 ns … 34.553 ns ┊ GC (min … max): 0.00% … 0.00%
Time (median): 4.707 ns ┊ GC (median): 0.00%
Time (mean ± σ): 4.770 ns ± 0.865 ns ┊ GC (mean ± σ): 0.00% ± 0.00%
Memory estimate: 0 bytes, allocs estimate: 0.

[mateuszbaran]

Thanks, looks good 👍 .