Many of the tests are brittle when adding new features to the SimHash

[thomasdullien]

This was a desired feature initially to make sure the underlying disassembly is good,
but makes tweaking / improving / adding new features difficult without breaking the
tests.

Particular culprits (and currently broken) are:
flowgraphwithinstructions.creategraph
flowgraphwithinstructions.parsejson
functionsimhash.zero_weight_for_mnemonics

It would be great if we could come up with a way that does not break quite so easily ...

[thomasdullien]

At least the first two may be doable by switching to a string comparison on the JSON serialized form?