test case for crazy resolvePath bug

[Rich-Harris]

No fix yet, but a test case for a very niche bug I've encountered. I think what's happening is that the same module is identified two different ways if it's imported by a module that's relative to base versus a module with an absolute path...

[Rich-Harris]

This PR now fixes the issue. It involved a spot of refactoring: rather than keeping a moduleLookup and externalModuleLookup object and passing it around, and using import id properties as lookup keys - inherently prone to this sort of bug - the objects representing imports are linked to the module objects themselves. Personally I think it's a little easier to follow the code as a result, though I stopped short of doing a larger scale refactor, as tempting as it was.