Add hooks, fickling.load(), and a JSON output format for usability

[suhacker1]

This PR makes multiple feature additions for usability. Specifically, this PR adds a fickling import hook, global function hook, fickling.load() function, and a JSON output format for the check_safety component of the CLI. Each of these features can make it easier to integrate fickling into different codebases and tools.

This PR also updates the examples and tests to reflect these new features. Additional important changes made include:

• Syncing is_likely_safe in fickle.py with check_safety in analysis.py: A new check_safety method was added to fickle.py as a wrapper with is_likely_safe being marked for deprecation.
• Refactoring analysis.py: Not only was ProtoAnalysis split for simplicity, but more structure was added throughout the different analysis classes to enable the reporting of detailed results.
• Adding new methods to pytorch.py: The PyTorchModelWrapper class now reports the identified file formats from the validation method.

I would especially appreciate feedback on:

• Whether the hook tailored for torch.load should be included
• Whether the naming throughout is consistent and appropriate
• How usable the interfaces now exposed by fickling are
• How useful the current state of the JSON output is

Example JSON Output ("test_unused_variables.json"):

{
    "severity": "OVERTLY_MALICIOUS",
    "analysis": "Call to `eval(b'[5, 6, 7, 8]')` is almost certainly evidence of a malicious pickle file\nVariable `_var0` is assigned value `eval(b'[5, 6, 7, 8]')` but unused afterward; this is suspicious and indicative of a malicious pickle file",
    "detailed_results": {
        "AnalysisResult": {
            "OvertlyBadEval": "eval(b'[5, 6, 7, 8]')",
            "UnusedVariables": [
                "_var0",
                "eval(b'[5, 6, 7, 8]')"
            ]
        }
    }
}

[Boyan-MILANOV]

@suhacker1 good job on this! It needs some restructuring but the core idea is there :) I've left a bunch of comments, some are nits, but most of them are about code and architecture. Let's address all of them and then I'll make a second pass on the PR.

[suhacker1]

Note: We decided not to include the PyTorch global hook in this PR. We also decided to remove the UNKNOWN severity type as we felt it was redundant.

[suhacker1]

@Boyan-MILANOV This is ready for another review!

[Boyan-MILANOV]

LGTM now! Good job! 🚀