Never fail to indent

[Hi-Angel]

A proper indentation engine should never look past the previous line. This is because a user frequently wants custom indentation, in all programming languages. A PureScript example:

f1 x = case x of
  true -> true
  _ -> false

f2 x = case x of
        true -> true
        _ -> false

So the proper behavior is: "take previous-line indentation, then if previous line has a modifier increase/decrease it". This is both simple, robust, typically meets the user expectations, and is AFAIK what all built-in major modes do.

Unfortunately we're long past that point, the purescript-mode indentation engine parses a lot of stuff it shouldn't. And frequently it gives a "parsing error". In my experience all those "errors" are bugs because the PS code is perfectly valid, but disregarding if it's a bug or an actual unfinished code in the buffer, engine should never leave a user with no indentation whatsoever.

So what this commit does is it detects such "errors", and takes previous indentation level. Even if it's not the correct indentation, in practice it's only off by purescript-indentation-left-offset, so a user often has much less to type compared to "no indentation" at all.

---

Now, it turns out the mode has another bug. As explained in the function being added, purescript-newline-and-indent calls indentation calculations on the wrong line. This isn't some special code that only purescript-newline-and-indent uses, it is a code that's being called by the TAB indentation as well.

As there are huge amounts of poorly documented code that shouldn't even be there in the first place, and are sometimes using local variables from other functions, I decided instead of trying to fix the code it would be more productive to just detect whether we're being called from purescript-newline-and-indent or not.