Clean up escaping

[travisbrown]

There are currently many different string encodings with different escapings, with lots of transition points between them. These transitions are currently handled in a fairly ad-hoc way, with escaping done at each of them as needed to get tests to pass.

I don't know of any specific bugs in this respect at the moment, but I'm sure there are some in there, and anyway the current situation won't be fun to maintain in the longer term. I need to work through exactly what needs to be escaped where, and to clean up the transitions. This should be relatively easy given the tests we have now.