speed up RodasTableau constructors

[oscardssmith]

fixes #2710.
Turns out hvncat is much more expensive than it should be/isn't constant folded.
Before:

julia> @benchmark OrdinaryDiffEqRosenbrock.Rodas4PTableau(Float64, Float64)
BenchmarkTools.Trial: 10000 samples with 4 evaluations per sample.
 Range (min … max):   6.920 μs …   7.793 ms  ┊ GC (min … max):  0.00% … 99.63%
 Time  (median):     11.370 μs               ┊ GC (median):     0.00%
 Time  (mean ± σ):   13.247 μs ± 105.500 μs  ┊ GC (mean ± σ):  11.22% ±  1.41%

                ▂▆███▆▂▁                                        
  ▃█▇▆▄▂▂▂▁▁▁▁▂▆█████████▆▅▄▄▃▃▂▂▂▂▂▂▂▂▂▂▂▂▂▂▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▃
  6.92 μs         Histogram: frequency by time         21.5 μs <
 Memory estimate: 22.20 KiB, allocs estimate: 142.

After

julia> @benchmark OrdinaryDiffEqRosenbrock.Rodas4PTableau(Float64, Float64)
BenchmarkTools.Trial: 10000 samples with 1000 evaluations per sample.
 Range (min … max):  1.939 ns … 11.273 ns  ┊ GC (min … max): 0.00% … 0.00%
 Time  (median):     2.219 ns              ┊ GC (median):    0.00%
 Time  (mean ± σ):   2.208 ns ±  0.198 ns  ┊ GC (mean ± σ):  0.00% ± 0.00%

                    ▂                             █▂          
  ▁▂▄▂▁▁▁▂▂▁▁▁▃▃▁▁▁▁██▂▁▁▁▁▃▂▂▂▃▆▁▁▁▁▂▃▂▁▁▂▄▄▂▁▁▁▆██▂▂▁▁▂▃▄▂ ▂
  1.94 ns        Histogram: frequency by time        2.38 ns <

 Memory estimate: 0 bytes, allocs estimate: 0.

julia> @benchmark OrdinaryDiffEqRosenbrock.Rodas4PTableau(Float32, Float32)
BenchmarkTools.Trial: 10000 samples with 909 evaluations per sample.
 Range (min … max):  124.638 ns …   1.764 μs  ┊ GC (min … max):  0.00% … 87.86%
 Time  (median):     139.304 ns               ┊ GC (median):     0.00%
 Time  (mean ± σ):   169.512 ns ± 179.489 ns  ┊ GC (mean ± σ):  15.16% ± 12.63%

  █▅▃                                                           ▁
  ████▇▇▇▅▃▅▅▅▅▃▃▃▁▃▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▃▃▄▃▁▅▅▆▄▆▇▇█ █
  125 ns        Histogram: log(frequency) by time       1.38 μs <

 Memory estimate: 704 bytes, allocs estimate: 10.

[oscardssmith]

note that as of Julia 1.12, this probably won't be nearly as important due to JuliaLang/julia#39729, but might as well write it this way anyway...

[oscardssmith]

Any idea if https://github.com/SciML/OrdinaryDiffEq.jl/actions/runs/15029774249/job/42239283382?pr=2712 cxould be related? I have a hard time imagining this changing anything...

[ChrisRackauckas]

No that's on master and I haven't had time to track down what that is.