Fixes complex range broadcasting with another range

[gbaraldi]

Fixes #32792.
To fix it I had to add precision to complex types, I can split it into two different PRs if necessary

[vtjnash]

How is it we end up attempting to create a TwicePrecision{Complex} instead of a Complex{TwicePrecision}? Is this because we refused to make TwicePrecision a Number?

[gbaraldi]

I have no idea. In the TwicePrecision code there is nothing related to complex numbers, so I think it works because of julia magic, until it gets to methods that specifically use IEEEFloat when it gives the method error in the issue

Also this fix needs a change, because when building, when it reaches twiceprecision it doesn't know what a complex number is and crashes

[mcabbott]

Is it possible just to make sure complex numbers are never used? range(1, 2+0im; length=3) does not make a StepRangeLen, and so perhaps λ .* xx should also only make one for real λ, and xx .+ yy should be more careful when promoting.

[mbauman]

I suspect this has the same problem as #55113 — it successfully constructs the object but will perform erroneous arithmetic (in particular division) with the resulting thing.