Brainstorm optimization

[JasonShin]

It seems like the emscripten version of libsvm runs into out of memory issue due to the poor memory optimization. The ticket aims to perform an initial brainstorming on how to make this better.

Iteration 1:
We can provide a few different variations of the API including

• Emscripten compiled libsvm; feature rich and runs on the browser, however it may have some memory optimization issues
• Native interfaced libsvm, which is only usable in Node environment
• Custom written SVM, not as feature rich as Emscripten libsvm but it is well optimized for large dataset

[JasonShin]

We can also look into how to optimize the performance of the generated code in Emscripten. I would greatly appreciate some help on this.