Excluding Big files from Machine Learning samples

In order to reduce Machine Learning samples zip size (from more than 2 Gb to 87 Mb), just commit this Pull Request:

https://github.com/dotnet/machinelearning-samples/pull/949

I added also a VB branch with many samples, taken from this repository and updated:

https://github.com/Nukepayload2/machinelearning-samples/tree/master/samples/visualbasic

In the VB branch I included functional tests to download all datasets, and to test all samples, in one click!

I think it is still possible to commit first the C# and F# sample repository, and then after to update Nukepayload2 VB repository and commit the VB branch to the Nukepayload2 VB repository, via a second Pull Request.

I failed to compile this demo (30/87 Mb), I think packages can be downloaded too, like other samples:
csharp\end-to-end-apps\Unity-HelloMLNET\HelloMLNET

This one compile but does not work:
csharp\getting-started\DeepLearning_ImageClassification_Binary

I did not test these demos:
csharp\end-to-end-apps\Scalable*

All other demos work fine (C#, F# and VB).
All packages are up to date (F# and VB).

Why committing directly to the master branch? At first I wanted to be able to check the size of the repository at each step, but finally, I only checked the size of the downloaded zip (It's almost impossible to delete a large file or a sensitive file from a repository, without causing an inextricable mess, especially for forked repositories), but I probably could have made a branch indeed.

Start by reviewing pull request #949 and the sample paths named in the issue, especially csharp/end-to-end-apps/Unity-HelloMLNET/HelloMLNET and csharp/getting-started/DeepLearning_ImageClassification_Binary. Check the listed Scalable samples and the repository zip size; done means large files are excluded, the affected samples build or run as described, and the reduced archive is validated.

Written by the indexing model from the issue text.