small fix for static array onehot

[ExpandingMan]

This fixes what I think was essentially a typo in the StaticArrays extension and adds some tests to check for it. Think the only reason it wasn't caught was because it was occurring only in the 3-argument version.

[ExpandingMan]

From the looks of it I assume these failures are normal. Locally I tested x86_64 linux on 1.10 which passes here.

[wsmoses]

hm 1.6/7/8 arent supposed to fail here.

Error is

Gradient & NamedTuples |    8      8
Gradient & SparseArrays / StaticArrays: Test Failed at /home/runner/work/Enzyme.jl/Enzyme.jl/test/runtests.jl:2779
  Expression: eltype(y) ≡ SVector{3, Float64}
   Evaluated: MVector{3, Float64} ≡ SVector{3, Float64}
Stacktrace:
 [1] macro expansion
   @ ~/work/Enzyme.jl/Enzyme.jl/test/runtests.jl:2779 [inlined]
 [2] macro expansion
   @ /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.6/Test/src/Test.jl:1151 [inlined]
 [3] top-level scope
   @ ~/work/Enzyme.jl/Enzyme.jl/test/runtests.jl:2762
Gradient & SparseArrays / StaticArrays: Test Failed at /home/runner/work/Enzyme.jl/Enzyme.jl/test/runtests.jl:2786
  Expression: eltype(y) ≡ SVector{3, Float64}
   Evaluated: MVector{3, Float64} ≡ SVector{3, Float64}
Stacktrace:
 [1] macro expansion
   @ ~/work/Enzyme.jl/Enzyme.jl/test/runtests.jl:2786 [inlined]
 [2] macro expansion
   @ /buildworker/worker/package_linux64/build/usr/share/julia/stdlib/v1.6/Test/src/Test.jl:1151 [inlined]
 [3] top-level scope
   @ ~/work/Enzyme.jl/Enzyme.jl/test/runtests.jl:2762
Test Summary:                          | Pass  Fail  Total
Gradient & SparseArrays / StaticArrays |   13     2     15
ERROR: LoadError: Some tests did not pass: 13 passed, 2 failed, 0 errored, 0 broken.
in expression starting at /home/runner/work/Enzyme.jl/Enzyme.jl/test/runtests.jl:2761
Package Enzyme errored during testing

[ExpandingMan]

Woah, that's incredibly strange considering that it's the result of calling SArray...

[ExpandingMan]

Well, I don't know what in the world is going on there, but considering this wasn't tested before I assume whatever the issue is was already present. I've weakened the tests so they shouldn't fail on older versions, so hopefully it wouldn't have to hold up this PR which is essentially just a bug fix.