[Epic] Improve the performance and quality of validation testing

[shumway]

We're consolidating our testing utilities across CK and CK Tile to simplify the infrastructure and promote best practices. This work focuses on performing all tensor initialization and validation on the GPU, using better numerics in validation, enabling more flexible reference calculations, and establishing simple testing patterns that reduce boilerplate. The result will be higher quality tests that execute much faster.

We will be collecting updates for several ongoing efforts, recording milestones, and outlining future work in this ticket.