Misc. improvements for GPU testing

• Draw functions w/ colors per-device in graphviz visualization.
• Fix bugs in GCM and loop nesting to allow forks to get to codegen.
• Add GPU schedules for dot and matmul tests that contain forks.
  • Currently commented out in the build scripts, just switch which schedule_in_src line is commented out in both to actually use the GPU schedule when using --features=cuda.

requested review from @prathi3

assigned to @rarbore2

mentioned in commit de78461b

merged