Ark: Completions in a debug session should reflect the debug environment

[DavisVaughan]

Right now, if you do debugonce(fn) and step into fn(), then the Debug Variables pane will show you variables that are relevant to that debug environment, but if you start typing in the Console then you will see completions that are relevant for the global environment rather than your debug environment.

This is obviously not correct, for a few reasons:

• It won't complete any variables you have in the function you are debugging
• If you happen to mask a global env function with a debug env object, like match <- 1:5, then our autocompletion engine will help you complete match(), with the parens, because it thinks you are in the global env
• If you try and debug our internal functions, like .ps.help.showHelpTopic(), which run in a child of the base env, then you will get completions for things like str() even though that is from utils and you can't actually even call it without utils::str()

We do capture the environment of the debug frame, so we've got that information to pass on to the completion engine. The tough part is plumbing it over there.

We probably need some central location of session level information that allows for updating bits and pieces of the session level information stored there - like the active console environment, which could then be queried by other parts of the code, like the completion engine (or even the standard Variables pane, but right now we've decided that that should always show the global env)

[jennybc]

My recent puzzlement in the debugger, while working on .ps.help.showHelpTopic() is part of what inspired this. To add a bit more color, I think what's so baffling about the fact that the LSP helps with completions that cannot then be evaluated is that I think most users assume that "R helped me complete this thing, therefore this thing is in scope". Which the current situation violates.