I've made some changes to the kernancompiler module and associated kernan features.
Firstly I reduced boilerplate and duplicate code, this required me to change ParseNodeMeta to actually provide a GraceObject and not a .NET Dictionary.

Secondly I added two extra methods to the kernancompiler module:

1. parse(n, c), this is like parse(c), but it tells kernan that the code came from the module called n, this will make parse errors more usefull (by telling the user which file has said error).
2. readGraceModule(m) this returns the contents of the grace module identifier by m, this performs roughly the same logic as kernan internally does for import "m" as foo; except that it won't treat resource files and .dll files specially. Thus you can say have:

foo.dll
foo.grace

And readGraceModule("foo") will return the contents of the .grace file, whereas an import "foo" will actually try and load the .dll file; the intention is that the .grace file will contain meta-data (like method specifications) for the .dll file.

In order to not duplicate code, I have modified kernan's import logic slightly to expose a TryReadModuleFile method that tries to read a module file, and instead of actually loading it (like the previous tryLoadModuleFile method did), it now just returns the contents of the module file as a string.

In order to minimise the number of changes made by this pull request, I have not used my cool Callback.cs library, but I still think it's a good idea.