`Console::get_mut()` should use `CONSOLE.try_with()`

I just ran into https://github.com/posit-dev/ark/pull/1222/changes/b6e980d5efb7f4ce865bfc8cd58792b6da6139bc with this panic message

cannot access a Thread Local Storage value during or after destruction: AccessError

What happened was:

• CONSOLE was destructing as R shut down
• CONSOLE.pending_futures was destructed via CONSOLE destructing
• One of the pending_futures was created with capture support, i.e. with ConsoleOutputCapture enabled
• ConsoleOutputCapture::drop was run as the future destructed. This tried to re-access CONSOLE via Console::get_mut(). But CONSOLE is destructing! So we get a panic.

We should analyze this and figure out the most robust solution.

For my use case, I was accidentally holding a permanently running RTask with capture support, so the easy solution for me was to just turn off capture support, avoiding ConsoleOutputCapture::drop from running.

But I think in theory that even short running futures with capture support that just so happen to be in pending_futures when R restarts and CONSOLE destructs can cause this to happen.

So maybe we need to be able to call something like Console::try_get_mut() from this location, which would use CONSOLE.try_with() instead of CONSOLE.with_borrowed_mut(). I think you can build a form of try_with_borrowed_mut() by looking at the definition of with_borrowed_mut() and replacing the .with() with .try_with().

Start with Console::get_mut(), with_borrowed_mut(), and ConsoleOutputCapture::drop, then trace how pending_futures are destroyed while CONSOLE is shutting down. Reproduce or test the capture-enabled future cleanup path and ensure destruction does not panic when CONSOLE is unavailable.