Yank ForwardDiff 0.10.33

[mcabbott]

This turns out to cause more disruption than intended, so perhaps it should be 0.11 not a patch version.

The offending PR is JuliaDiff/ForwardDiff.jl#481

Some reports of issues aren't issues, but the ones that are are:
JuliaDiff/ForwardDiff.jl#606
JuliaDiff/ForwardDiff.jl#607
JuliaDiff/ForwardDiff.jl#609

[oxinabox]

I think this case it would have been better to just revert and release 0.10.34, without yanking 0.10.33

Like all patch releases, (though not true pre-1.0.0) the older version had bugs.
But we don't yank all penultimate patch versions. (Though I have suggested that Pkg should by default treat them that way, but that's another dicussion)

The particular thing that yanking does that just making a new release doesn't is it means you can do ]add --preserve=all Foo for minimal changes to a known working system. (nor update ---preserve=all Foo in 1.9 I suspect).
And this broke stuff for Invenia's workflow trying to make minimal changes to prod.

[mcabbott]

Well IDK, there appear to be many opinions. Want to make a branch for 0.10 versions & tag that as 0.10.34?

Reverting things seems to be a way to make sure they rot.

[oxinabox]

revert release patch, then unrevert and release breaking IMO is best

Well IDK, there appear to be many opinions. Want to make a branch for 0.10 versions & tag that as 0.10.34?

It wouldn't help. The problem is the yanking of 0.10.33 being more disruptive than just releasing a 0.10.34 that is identical to 0.10.32.