Fix extra prompting and manual debugger commands

[andyleejordan]

This requires that we set the REPL to exit correctly. Note that we don't
have a notion of "disconnect" so we removed that logic. We are assuming
the debugger is "active" a little earlier now, that is, also when we've
received an LSP notification to start it (via a launch or attach
handler). Because we receive a notification on "disconnect" even for a
script just finishing, we need to know that the server is not active and
so not cancel the current prompt.

[andyleejordan]

@SeeminglyScience Just added the commit that removes the WriteWithPrompt workaround from the andschwa/improvements branch (and originally explored in #1769). It's just cleaner now, the first commit is obviously the actual fix. Now the existing prompt gets used correctly by whatever we're about to run.

[andyleejordan]

Kind of funny cosmetic bug: if you run a script, and then type input at the next prompt, and then run the script again, then PSReadLine doesn't save that input. But if you've invoked anything else in between (or type that text before running the script) it works fine. Strange, we'll fix it later.

[SeeminglyScience]

LGTM but one thing. HMU if you want to keep it, doing request changes to not trigger auto merge