Fix reparse trailing escape

[alexrutar]

Two changes here:

First, fixes #66 to check for a trailing unescaped \.

The other fix, which is in the second commit and which I am slightly less sure about, is to replace an ASCII whitespace check with a is_whitespace check in Atom::new. This behaviour was previously changed in #65 for Pattern::(re)parse, is it also correct for this to take place in Atom::new? If this is incorrect I'll drop the second commit.

[pascalkuthe]

Makes sense, thanks!

[pascalkuthe]

Ah actually the test seems to be falling

[alexrutar]

Ok, great!

There is another bug still: escaping is for some reason not handled correctly at all in the presence of Unicode, e.g. Atom::parse("foö bar"). Will fix it right now and then force-push.

[alexrutar]

Oops, force-pushed just as you added a review, maybe it was on the thing that I fixed in any case.

[alexrutar]

Ok everything seems to be fine now

[alexrutar]

Oh of course this is still is still wrong in the ASCII case now too and needs to be fixed there too. Sorry about that, should have read the surrounding code more carefully before pushing the PR!

[alexrutar]

Now the parsing code is changed even more. I added some tests with all of the edge cases that I found in the previous implementation, so hopefully the new one is more correct, but to be honest with manual parsing code like this without extremely good test coverage I am always a bit concerned.

1. In the Unicode case, there was a \ push that had to be deferred until the end beginning of the subsequent loop, since it was causing incorrect handling of scapes in the presence of Unicode. This was causing invalid handling of e.g. foö\ bar.
2. I've re-implemented the ASCII case with essentially the same code as the Unicode case, but with the usual ASCII shortcuts.