Libmjollnir core algorithm should be recursive

[may]

Currently we do a linear read of the binary code for constructing the control flow graph, this is not good for multiple reasons:

- we cant do local analysis given an entry point (we HAVE to analyse a whole section at all the time, and it does not scale when analysing the kernel).

- we cant analyse efficiently the obfusfated code that jumps in the middle of instructions. In the new algorithm, an instruction (its bytes) can potentially be present in multiple basic blocks if this case happens.

Instead we should use an entry point and a max depth, and follow the control flow edge when constructing the CFG.