Reinstate handlers during Rprofile sourcing

[thomasp85]

Fix #1101

This PR ensures that our error, warning, and message handlers are all in place during startup-sourcing of .Rprofile etc.

It does this by calling initialize_errors() prior to sourcing which re-registers our handlers. It was necessary to modify the logic somewhat to avoid very noisy

pushing duplicate `error` handler on top of the stack
pushing duplicate `warning` handler on top of the stack
pushing duplicate `message` handler on top of the stack
pushing duplicate `interrupt` handler on top of the stack

during startup, so we now filter the existing handlers before applying our own (again). Trying to surpress this by wrapping in suppressWarnings() etc results in system crash.

To test this PR add this to you .Rprofile

message("Hello")
globalCallingHandler(test = function(...) print("test"))

and check that "Hello" is printed in white, not red, and that a "test" handler has been correctly registered

[lionel-]

It would be nice to add some integration tests here (e.g. look at https://github.com/posit-dev/ark/blob/main/crates/ark/tests/kernel-r-profile.rs)

[lionel-]

ark rather than ps

[lionel-]

registered_ark_handlers

"registered" is clearer to me, and plural makes it clearer this is a vector

[lionel-]

What is preventing usage of identical() here?

[lionel-]

This is not exported so shouldn't use .ps. prefix.

[lionel-]

We'll need a comment in initialize_errors() to mention that particular calling context.

Are we certain that no ill side effect comes from reregistering the global handlers in that top-level context? Does R restore the handler stack to the previous state? (It might very well do that).

To make things clearer I think I'd prefer not using initialize_errors() which in my mind should only be called once per file. Probably best to split into initialize_global_handlers().