Right now the generated code will link to libjsonschema to get the encoding API (mostly jsch_decoder and jsch_encoder). We could avoid this by embedding the low-level encoder/decoder into the generated file directly.

This raises a few questions:

• Cannot update the low-level encoder/decoder without re-generating the code, but IMHO that's fine.
• We need to make sure a single binary can link multiple generated C files from different schema files. We can make the low-level encoding API purely file-local via static.
• How do we expose functions like jsch_encoder_create?