[idan] Improve Bidi Rules implementation

* As the first paragraph of The Bidi Rules section explains, the rules
need to be ignored if there are no Bidi labels present in the domain
name. So, add `is_bidi_domain` evaluation to `processing()`, and pass it
down to `passes_bidi()` to act on.

* Add unit tests for the bidi rules, making it faster and easier to
maintain the feature.

* The new data file depends on the GC of `U+11C3A` to be `Mark`, but
`is_combining_mark()` returns false on this character. Since the
character has `GC=Mark` in Unicode 9.0, and the published version of
`unicode-normalization` is also using Unicode 9.0 data, there must be a
bug somewhere in the crate that needs to be fixed. To fix the problem
locally until there's an upstream fix, I've added another check, on top
of `is_combining_mark()`, using `Bidi_Class=NSM`, which covers two out
of three `GC=Mark` subcategories. This local fix allow conformance tests
to pass 100%.

Author: behnam

idna/Cargo.toml

@@ -20,5 +20,5 @@ rustc-serialize = "0.3"
 
 [dependencies]
 unicode-bidi = "0.3"
-unicode-normalization = "0.1.3"
+unicode-normalization = "0.1"
 matches = "0.1"