EPA Algorithm in C++, faster EPA CUDA

[vismaychuriwala]

This update adds a CPU EPA algorithm and parallelizes aspects of the GPU EPA versions, updating API for ease of use and python wrapper updates.

Features

• Added a new EPA algorithm in C++ in \scalar alongside the current GJK.
• A much faster (~10x) EPA kernel in CUDA using shared memory and parallelized face creation.
• Streamlined API for CUDA, no longer using separate arrays for witness points, updates witness points in simplex in place.
• New indexed API for CUDA EPA.
• Reworked GPU python wrapper to work primarily using indexed GPU batches (Allocate and upload once, query multiple times at different indices for both GJK and EPA).
• Updated tests to check consistency in CPU-GPU versions (GJK+EPA), added simple EPA tests.

[MattiaMontanari]

@vismaychuriwala nice API!

[MattiaMontanari]

Thank you @vismaychuriwala !