Breakage in upcoming Julia 1.7

[KristofferC]

The tests for this package fails on the upcoming 1.7

ERROR: LoadError: MethodError: no method matching generic_lufact!(::Matrix{Float64}, ::Val{true}; check=true)
Closest candidates are:
  generic_lufact!(::StridedMatrix{T}) where T at /workspace/srcdir/usr/share/julia/stdlib/v1.7/LinearAlgebra/src/lu.jl:136 got unsupported keyword argument "check"
  generic_lufact!(::StridedMatrix{T}, !Matched::Union{LinearAlgebra.NoPivot, LinearAlgebra.RowMaximum}; check) where T at /workspace/srcdir/usr/share/julia/stdlib/v1.7/LinearAlgebra/src/lu.jl:136
Stacktrace:
  [1] #lu!#2
    @ ~/.julia/packages/RecursiveFactorization/ec89K/src/lu.jl:12 [inlined]
  [2] lu! (repeats 2 times)
    @ ~/.julia/packages/RecursiveFactorization/ec89K/src/lu.jl:9 [inlined]
  [3] (::DefaultLinSolve)(x::Vector{Float64}, A::Matrix{Float64}, b::Vector{Float64}, update_matrix::Bool; reltol::Nothing, kwargs::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ DiffEqBase ~/.julia/packages/DiffEqBase/lULzQ/src/linear_nonlinear.jl:77
  [4] DefaultLinSolve
    @ ~/.julia/packages/DiffEqBase/lULzQ/src/linear_nonlinear.jl:63 [inlined]
...

Since generic_lufact! is an internal method, I believe this should be fixed in the package.

[chriselrod]

Implement our own generic_lufact! optimized for the size <= 10 case?

I added support for placing upper bounds on the sizes of dynamically sized loops, so it'd also be worth checking if this could solve the "avx slower for size <= 10" case.

[YingboMa]

I have a fix locally. I'll make a PR when it's ready.