Forward mode `gradient` doesn't support tuples with varying element types

[mhauru]

MWE:

julia> f(t) = t[1]*t[2]
f (generic function with 1 method)

julia> Enzyme.gradient(Enzyme.Forward, f, (1.0, Float32(2.0)))
ERROR: MethodError: no method matching onehot(::Tuple{Float64, Float32})

Closest candidates are:
  onehot(::Type{Tuple{Vararg{T, N}}}) where {T, N}
   @ Enzyme ~/projects/Enzyme.jl/src/Enzyme.jl:1557
  onehot(::Array)
   @ Enzyme ~/projects/Enzyme.jl/src/Enzyme.jl:1527
  onehot(::Array, ::Int64, ::Int64)
   @ Enzyme ~/projects/Enzyme.jl/src/Enzyme.jl:1531
  ...

Stacktrace:
 [1] macro expansion
   @ ~/projects/Enzyme.jl/src/Enzyme.jl:1833 [inlined]
 [2] create_shadows
   @ ~/projects/Enzyme.jl/src/Enzyme.jl:1811 [inlined]
 [3] gradient(::EnzymeCore.ForwardMode{false, EnzymeCore.FFIABI, false, false}, ::typeof(f), ::Tuple{Float64, Float32})
   @ Enzyme ~/projects/Enzyme.jl/src/Enzyme.jl:1970
 [4] top-level scope
   @ REPL[97]:1

Looking at the definition of onehot(::Type{NTuple{N,T}}), it doesn't seem to use the parameter T anywhere. Could it just be dropped?

[wsmoses]

fixed in #2574