feat: Initial Support for FlatBuffer context

[lokitoth]

Add support for using flatbuffer spans as input to RLClientLib when using a VW model.

Note: Currently only working in CB models when using choose_rank().

• [x] Fix memory leak when providing bad buffer (e.g. JSON) to a Flatbuffer-configured LiveModel
• [x] Figure out why the build breaking on fb_parser_test
  • reinforcement_learning builds on ubuntu1804; vowpal_wabbit builds on ubuntu2004
• [ ] Verify whether CA models work out of the box with Flatbuffer contexts
• [ ] When no more changes to VW need to be made, return submodule back to pointing to a trunk commit of VW

Stretch:

• [ ] Support RequestDecision/MultislotDecision (need to implement context introspection for FB contexts)
• [ ] (unlikely) Support Episodic (need to implement history generation / merge; to do this right requires a fair bit of engineering)