Define outer repeat method for TracedRArray

[AntonOresten]

Closes #360

This PR leverages broadcast_to_size to define outer repeating (the default for Base.repeat) of traced arrays.

I'm new here, so I may have missed something, but this looks promising:

julia> @code_hlo repeat(rand(2,3) |> Reactant.to_rarray, 5, 7, 11)
module {
  func.func @main(%arg0: tensor<3x2xf64>) -> tensor<11x21x10xf64> {
    %0 = stablehlo.reshape %arg0 : (tensor<3x2xf64>) -> tensor<1x1x1x3x1x2xf64>
    %1 = stablehlo.transpose %0, dims = [5, 4, 3, 2, 1, 0] : (tensor<1x1x1x3x1x2xf64>) -> tensor<2x1x3x1x1x1xf64>
    %2 = stablehlo.broadcast_in_dim %1, dims = [0, 1, 2, 3, 4, 5] : (tensor<2x1x3x1x1x1xf64>) -> tensor<2x5x3x7x1x11xf64>
    %3 = stablehlo.transpose %2, dims = [5, 4, 3, 2, 1, 0] : (tensor<2x5x3x7x1x11xf64>) -> tensor<11x1x7x3x5x2xf64>
    %4 = stablehlo.reshape %3 : (tensor<11x1x7x3x5x2xf64>) -> tensor<11x21x10xf64>
    return %4 : tensor<11x21x10xf64>
  }
}

Here's a testset (wasn't sure where to put it) that takes a few cases and compares the results to the generic method from Base, which passes for me.

@testset "repeat" begin
    for A_size in [(2,), (2,3), (2,3,4), (2,3,4,5)]
        for counts in [(), (1,), (2,), (2,1), (1,2), (2,2), (2,2,2), (1,1,1,1,1)]
            A = rand(A_size...)
            A_ra = Reactant.to_rarray(A)
            @test (@jit repeat(A_ra, counts...)) == repeat(A, counts...)
        end
    end
end

[wsmoses]

perhaps in tests/basic.jl? regardless, I'll let @avik-pal review being more familiar with repeat semantics than I