Secml-malware

Amazing content in this new release:
* added support to Nevergrad optimizer to create faster attacks
* added support to ping VirusTotal with samples through the CClassifierRemote
* added MalConv with 2MB of input window, with pre-trained weights loaded from open source repositories (more details in the code)

What a time to be a SecML-Malware user 🚀

**Full Changelog**: https://github.com/pralab/secml_malware/compare/v0.3.1...v0.4

Fixed bugs in 0.2.9, added larger MalConv model

Since EMBER has an issue with latest numpy, while I'll wait for the pull request to merge, I have used my backup fork to fix some bugs.

Fixed bug on Kreuk attack implementation.

Finally, the SOREL DNN is supported when using benign/malware scores!