Better interactive backward-kill-word

[r-zip]

This PR implements a fix to backward-kill-word in interactive-haskell-mode.

The way things are currently implemented, calling backward-kill-word in the REPL causes the point to move to the beginning of the line if there is nothing else to delete on the line. The user then gets a message that the prompt text is read-only.

This is annoying for two reasons:

1. The message about read-only prompt text clutters the minibuffer, and
2. The user has to move the point past the prompt again to continue typing.

The changes in this PR improve the ergonomics of interactive-haskell-mode by detecting when backward-kill-word would move past the prompt. In that case, the point is just moved to right after the prompt.

I've chosen to re-bind backward-kill-word in the haskell-interactive-mode-map, and from what I can tell, it works.

I've never contributed a PR to an open-source project, but I've read the wiki page on contribution guidelines. The implementation has gotten more complex since I originally opened the PR, so if you guys want unit tests, I can write those.

Old behavior:
https://asciinema.org/a/OYRtYHCAUSNZjWq2NoCn4UEdb

New behavior:
https://asciinema.org/a/fm3Gd6WV1Q7IMWMzAxwcNQTCV

[purcell]

Thanks for taking the time to submit this! I feel like there must be a more general way to tackle this, since many other commands exist which would need to be overridden in this way, e.g. backward-kill-sexp, kill-back-to-indentation etc.