* Expose skippy backend device survey
* Bundle GPU benchmark backends
* Fix CI benchmark wiring
* Allow manual Docker CI runs
* Restore CPU-only runtime capacity
* Fix CUDA loading issue on Linux
* add debug information to help trace issues on CUDA devices with two GPUs
* fix linters / tests
* fixups for remaining tests via mutex
* fix ROCm build
* coupel more fixes for windows
* debugging on ci is cool
* fix remaining red tests after rebase
---------
Co-authored-by: Nick DiZazzo <nick.dizazzo@gmail.com>
