Tmp/recheck func

[JernKunpittaya]

Revisit and fix zkstats lib

-param_visibility in settings: change from 'private' to 'fixed' to enforce exact same onnx file for prover and verifier
-introduce precal_witness as a file to store all precalculated witness from prover (i.e. the precalculated result & other intermediates), then when the verifier generates their model, it just hardcodes these precalculated_witness into it.
-With reasons above, we change the flow to be as shown in Readme and all examples in 'examples' folder.
-Remove some uses of torch.tensor() since this basically ignore everything inside and treat it as just a constant. So, can only use it when variable inside () is a constant.
-Delete Where class since we wont include where as precalculated_witness since that will leak too much information
-MagicNumber uses 99.999 instead of 9999999 to avoid blow up
-Make sure our op.result can be transformed into keras
-Make sure computation_to_model always return bool, result format even though we dont use State
-Make default value argument make sense as None: op_dict in ops.py never changes value inside ops.py