Reduce memory allocation when squaring a nested Taylor1

[LuEdRaMo]

In master, running the following block of code:

begin
    using TaylorSeries
    import Random
    Random.seed!(1234)
    inorder = 6
    outorder = 25
    x = Taylor1([Taylor1(rand(inorder+1), inorder) for _ in 1:outorder+1], outorder)
    y, z = zero(x), zero(x)
    for r in [0, 1, 2, 3, 4]
        println("x^$r")
        for k in eachindex(z)
            TS.pow!(z, x, y, r, k)
        end
        @time for k in eachindex(z)
            TS.pow!(z, x, y, r, k)
        end
        println("|z - x^$r|∞ = ", norm(z - x^r, Inf))
    end
 end

produces the following output:

x^0
  0.000005 seconds (35 allocations: 1.219 KiB)
|z - x^0|∞ = 0.0
x^1
  0.000005 seconds (35 allocations: 1.219 KiB)
|z - x^1|∞ = 0.0
x^2
  0.000034 seconds (60 allocations: 3.953 KiB)
|z - x^2|∞ = 0.0
x^3
  0.000059 seconds (35 allocations: 1.219 KiB)
|z - x^3|∞ = 0.0
x^4
  0.000041 seconds (35 allocations: 1.219 KiB)
|z - x^4|∞ = 0.0

showing that squaring a nested Taylor1 allocates twice the memory. This PR proposes a quick fix.

[lbenet]

Hve you checked that everything is ok in TaylorIntegration, in particular wrt the use of @taylorize?

[coveralls]

coverage: 92.733% (+0.1%) from 92.609%
when pulling e8e6a7d on LuEdRaMo:lerm/pow
into fc1dcbd on JuliaDiff:master.

[LuEdRaMo]

Hve you checked that everything is ok in TaylorIntegration, in particular wrt the use of @taylorize?

In my computer, all TaylorIntegration tests pass except for those related to the DiffEq extension, which I reckon is not related to this PR.

[lbenet]

I've also noticed that on my computer; I think it is due to some recent upgrade in DiffEq, that I haven't yet traced back

[lbenet]

Incidentally, when you bump the patch version related to this PR, can you also include compatibility with IA v0.23 (so it works with 0.23 and 1 as well)? See this for some context.

[LuEdRaMo]

Incidentally, when you bump the patch version related to this PR, can you also include compatibility with IA v0.23 (so it works with 0.23 and 1 as well)? See this for some context.

Done.

If you have no further comments, I think this PR is ready...

[lbenet]

Thanks a lot for this improvement!