Add testcase and fix for indentation bug

[ghost]

Grammar::Tracer gets kinda confused if a traced .parse throws an exception followed by another trace in the same process.

This fix doesn't appear to have any unwanted side effects, though I haven't tried it with anything crazy.

[meisl]

There are some problems with this pr, in ascending order of severity (ie here goes "critique"):

• Debugger has the same problem so both should be fixed
• silently glossing over an exception or even only a failure ("unthrown exception", what you get with fail) is generally not a good idea. The more (not) so for a Tracer or Debugger - at least if it aims to deserve the name... (with this pr it looks like a simple failure to parse and does NOT give an indication of the exception)
• (confusion of) indentation is just one symptom (in Tracer it happens to be the only one) of a more general problem - which is not at all fixed by this pr

The real problem is: Both Tracer and Debugger have state like current call-depth (~> indent), declared / manual breakpoints, auto-continue/single-step etc. This, in itself is fine and need be. However, the question is with what to associate this state?. Currently it's associated with the meta-class, which means that there's only one state, for each and every grammar, and instance thereof, plus - even worse - one state only for the whole program's live-time!

This has several implications:

• subsequent parses start off with the state from the previous one (causing described "indentation bug")
• ...which in turn also makes (substantial, real) testing nearly impossible
• not reasonably useable with more than one grammar at a time, be it instances or classes, be it via inheritance or by one grammar calling another explicitly

So(lutions):

• state should be associated with the current Grammar invocant (class or instance, as the user chose)
• state should be reset whenever parse, subparse or parsefile is called:
  • auto-continue/single-step should be reset to False/True, resp.
  • stop-at-fail should be reset to False
  • stop-at-name should be reset to empty string
  • breakpoints set manually via bp add <name> should be all deleted
  • declared breakpoints (via is breakpoint or will break {...}) should be reset to what's been declared
• re exceptions thrown from within regexes: we might catch them and tell the user, then ask her what to do (well, what options are there, really?) OR just die, which very likely is what the exception was for in the first place.
• as for "Tracer's functionality is a subset of Debugger's" (ref first pt in critique): in order to handle this properly there's quite some refactoring needed. In the end there should be an "EventEmitter" (Observer pattern), or layers of such, to which Tracer and Debugger are just different subscribers that react differently to events. For the time being there'll be just enough to avoid dead-stupid copy&paste but not the full-blown Observer pattern.

I'll soon provide a pull-request implementing pr #7 implements pt 2 and - hence safely - pt 3, by just enough of pt 4. Plus a bit of preparation for pt 1. Once that's ready I think I'll move much of Maybe I should move this rationale over there or maybe into a new issue? (of course leaving a link here)

Any input/comment/advice/critique is very much welcomed, as always :)