Add naming collision strategy that uses XPATH prefices

I just figured why the unmarshal() method of BondNotioanl is wrong. Have a look at SourceFactory.createUnmarshalMethods(), l994ff. At the bottom of this method, a JSourceCode instance is created that holds the method body of the unmarshal() method to be created. As a mattr of fact, this is created before class nam econflict resolution. Whilst the toString() methods of e.g. JMethod and JMethodSignature are linked to a JClass instance (whose local name we change in the strategy implementation) and thus lazily create the code, this JSource instance is created too early.

Proposal: introduce something similar to JMethodBody, that takes an arbitrary number of parameters, and that has a toString() method.

Next problem: the field _notional in BondComplexType should have a type of BondNotional. Unfortunately, it does not .. .

Somehow I think we should switch towards an approach where we deal with those collisions when creating ClassInfo/FieldInfo instances. Given a new FieldInfo instance (with a name 'notional'), which we want to add to an existing ClassInfo instance, it should be easy to figure a path from this ClassInfo in question to the root class (classes generated for global types, who don't have a parent anymore). Given that path, one could maintain a registry of path to ClassInfo/FieldInfo mappings.

Completely reworked patch, adding a class name conflict resolution at the JClass level. With this patch, all CTF tests finish fine. I still have to clean out the old resolution strategy solution, as it really does not add any additional value anymore.

The new usage of a JClassRegistry is actually configurable via a new builder property.

New and improved patch, adding code to SG.generateAllClasses() to pre-register global group definitions as well.

Improved patch, where I removed some old (aka outdated) code from the original approach. The use of strategies has been simplified again to what it used to be earlier.

Updated patch, with some improvements in xPath calculation (taking types into account as well).

Updated patch, where I have added code to deal with local group references as well, by simply accepting the class name as suggested. This time, generating sources for the complete FpML schema set completes successfully.

Improved and final patch.