Create ML model to eliminate false-possitives and increase accuracy

[pdparchitect]

It does not matter how well we tune the regular expressions the method will always be subject to false-positives. One effective way to reduce the noise is to use an ML model for filtering (not detection).

To build the ML model, the following steps are required:

1. Download a large body of content known to produce false-positives (js files and other source code).
2. Run the current set of detectors to extract leaks (the generic secrets set is most suitable).
3. Use brain.js or an equivalent framework to train a model to spot the false-positives.
4. Compile the model.
5. Use the model to filter results from problematic detectors (again, the generic secrets is most suitable).