Add DRIVE tensor encoder

[amitport]

Adapted from the source published with "DRIVE: One-bit Distributed Mean Estimation" (NeurIPS '21; https://arxiv.org/pdf/2105.08339.pdf), and the algorithm presented therein.

Without going into the details of the paper, the key reason to include this as a core encoder is that for practical distributed scenarios, this encoder achieves accuracy results that are the same as using float32 while using only one bit per coordinate.