Fix to_vec with Integers

[willtebbutt]

Integers aren't considered differentiable, and they're certainly not perturbable. This bug fix enforces that.

This will probably break some things downstream that depend on this bug.

[oxinabox]

What was the previous behavour?

[mzgubic]

What was the previous behavour?

I think previously it just treated the integers as embedded in the reals and returned a vector of length 1 with integer transformed to a real.

This will probably break some things downstream that depend on this bug.

We might want to add tests for this as well, like in:
https://github.com/JuliaDiff/FiniteDifferences.jl/pull/189/files

Also, the _int2zero in that PR was to get around some really annoying behaviour where j'vp is not computed properly. The reason is the following situation:

julia> v1, b1 = to_vec(6)

julia> v2, b2 = to_vec(6)

julia> res = b1(v1 - v2) # b1 will just return 6 whatever the vector is
6

_int2zero didn't feel like the right solution at the time, and it still doesn't. Maybe there is a better way around this

[willtebbutt]

I think previously it just treated the integers as embedded in the reals and returned a vector of length 1 with integer transformed to a real.

Exactly.

We might want to add tests for this as well, like in:
https://github.com/JuliaDiff/FiniteDifferences.jl/pull/189/files

Lol sorry, I totally forgot about that PR. I'll close this PR in favour of that.

[mzgubic]

I am not working on it actively, feel free to take it over!