reject non-alphabet characters in base64Binary lex

[aizu-m]

Noticed while checking how xsd:base64Binary values get validated. A document whose base64Binary value carried trailing junk still validated clean:

<t:root>SGVsbG8=!!!!</t:root>   ->   doc.validate() == true

Traced it to JavaBase64Holder.lex. It decodes with Base64.getMimeDecoder().decode(v), and the JDK MIME decoder silently drops every character that is not in the base64 alphabet, not only line separators. So SGVsbG8=!!!! decodes to Hello, SGV!!!sbG8= decodes to Hello, and !!!! decodes to an empty array. Each is accepted instead of being reported invalid. lex is the validator the streaming Validator calls for base64Binary through validateLexical, and set_text calls it too, so an out-of-alphabet value passes document validation.

The sibling JavaHexBinaryHolder.lex does not have this problem: HexBin.decode returns null on any non-hex character and the value is reported invalid. base64 had drifted from that behaviour.

Fix scans the value first and reports it invalid when a character is neither in the base64 alphabet nor XML whitespace, then decodes as before. Whitespace and line-wrapped values are untouched, so valid input decodes exactly as it did.

before: <t:root>SGVsbG8=!!!!</t:root> validates
after:  reported invalid; "SGVsbG8=", "SGVs bG8=" and newline-wrapped values still validate

Regression test added in Base64BinaryValidateTest.