Implement VariadicAny around pg_sys::FunctionCallInfo

[workingjubilee]

The current FunctionCallInfo is unnecessarily unsafe when presented as an argument to a function. We could have a safe wrapper around it that only exposes sensical behaviors when it is presented as an argument to a function. This could also simplify everything we currently do... pg_getarg, etc.

[workingjubilee]

This is made particularly relevant because pgrx has a neat trick where, if the final argument is FunctionCallInfo, it will basically smuggle that in so that you can inspect pieces of the function call state. It might make the most sense to basically have an iterator of "the pieces of argument unpacking".

[workingjubilee]

This is no longer a "consideration": I have identified this as actually very important to moving forward, because pg_getarg has to be replaced. But I increasingly think pg_getarg itself should not be directly broken, as direct users of it do exist, and they assume the burden, when they call it, of verifying they are calling it soundly, as it is an unsafe fn. I want a safer interface to replace it that moves more of its safety obligations to its construction time.

[workingjubilee]

Also this would be useful for allowing a VariadicAny: #1505