impose stricter and earlier validation on ccall/cfunction types

[vtjnash]

Closes #43636

[NHDaly]

Cool, thanks Jameson! The changes seem sane to me, although i'm pretty unfamiliar with this code. The tests definitely seem reasonable. Thanks for making the change! 👍

[fingolfin]

This seems to have broken several packages that worked well so far. I am just wondering if that's intentional? From reading the issue, it sounded to me as if this was mostly there to fix things that would have crashed before, but in our case, we had a ccall with return value Function, and it worked perfectly fine so far. Now it is rejected. Is that as intended? If so, perhaps it deserves a mention in some NEWS item or whatever, because it'll break existing working code?

[vtjnash]

Function seems like a mildly odd choice for a return type there, where using Any agrees better with the ccall argument used to create it. You are right that it was intentional, and that it is more strict now than necessary. We have had the alternative also of Ref{Function} for a long time for promising "this returns a valid julia pointer to something that I promise is a subtype of Function".

[andreasnoack]

I've bisected this PR as the cause of an error similar to https://discourse.julialang.org/t/package-compiler-errors-with-system-image-too-large-message/85849. With version 1.7.x of Julia, our system image is about 1.1GB but after this PR, it fails with the system image too large error.