Convert LeNet-MNIST example to X10

[Andr0id100]

First of many PRs for #514.

Been busy with a few things lately hence the hiatus. Played around with X10 the last couple of days and I am really exited about the future of the S4TF project. Great work

[BradLarson]

As a quick update: I checked this out on Linux and it works great there, however, we're running into some issues with X10 on macOS that we're looking into now. Specifically, this example segfaults under X10 on macOS with both our 0.9 release toolchain and the current nightlies. It's not a problem with what you have here, but something internal.

Before we merge this in, I want to see if we can fix this. If not, we may need to add selective compilation checks to use the TF eager device by default in this example on macOS so that it still works out of the box. I'll let you know what we find.

[RahulBhalley]

Yup, I also experienced segmentation fault when executing on X10 on my macOS earlier.

[BradLarson]

Hope you don't mind, I pushed a modification to conditionally disable X10 support on macOS. We're not going to be able to fix the issue there with this model for a little bit, so it was best to get this in now.

Thanks for working on this, but before you move forward with other models, PR #586 will introduce a new and much cleaner training loop interface that will accommodate X10. I'll be working to migrate over our other models to that once that PR lands.