fix inverted xsd length facet check in anyURI and NOTATION holders

[aizu-m]

Spotted while reading the *HolderEx facet checks side by side. The length-facet branch in JavaUriHolderEx.check reads:

if (length == m) {
    return false;
}

It returns "invalid" when the length matches, the opposite of the minLength/maxLength branches right below it. JavaNotationHolderEx.check has the same defect, spelt as the double negative !(v.length() != m).

Both sit on the validate-on-set path (set_text -> check). I compiled an anyURI restricted with xs:length value="5" and set values with VALIDATE_ON_SET:

len5 (matches facet)      : XmlValueOutOfRangeException   <- valid value rejected
len3 (violates xs:length) : accepted                      <- facet bypassed

So a correctly sized value is thrown out, and a wrong-length one slips past the length check entirely. JavaStringHolderEx and the validate_simpleval path already test length != m, so I flipped both conditions to agree with them.

Added a regression test that fails on trunk and passes with the change.

[pjfanning]

merged with 568c4c6