Avoid extra philox calls with ElementRNG

[nhz2]

Fixes #721

This PR avoids some extra philox calls when using ElementRNG. ElementRNG is used as an AbstractRNG when sampling any non batched type including types with custom samplers defined in different packages.

This PR also makes ComplexF16 a batched type.

[nhz2]

There are many interesting bugs with using Float16 and Int128, but I don't think they are related to this PR.

[nhz2]

Due to JuliaLang/julia#62507 I'm restricting Float16 tests to Julia versions 1.13 or later.

[maleadt]

Due to JuliaLang/julia#62507 I'm restricting Float16 tests to Julia versions 1.13 or later.

That's excessive. Float16 had been working fine before; why does it only fail here? If it's specific to a test, you can always add a specific version-gated filter on top of the eltypes passed into the testsuite

[nhz2]

The Julia 1.12 miscompile only happens on CPUs with avx512_fp16. Github actions will sometimes use a CPU with avx512_fp16 now. If you run the existing tests enough times you can see the failure. I've limited the removal of Float16 tests to only Julia 1.12 and avx512_fp16 CPUs.

[maleadt]

The Float16 issue should be solved on 1.12.7, so I removed the workaround. Also rebased the PR.

[nhz2]

Thanks. The AMDGPU i128 problem will hopefully be fixed by either llvm/llvm-project#213523 or llvm/llvm-project#214518

[maleadt]

Thanks. The AMDGPU i128 problem will hopefully be fixed by either llvm/llvm-project#213523 or llvm/llvm-project#214518

In the interim, let's do JuliaGPU/AMDGPU.jl#1066