[pt2][inductor] only use global cache on MAST (#105375)
Summary:
until we can further investigate the autotuning differences between MAST and non-MAST (devserver) environments, turn off the global cache for all non-MAST environments. this ensures we don't see unexpected regressions
also update scuba logging for cache lookup, and add scuba logging for autotuning results.
Test Plan: sandcastle + CI
Differential Revision: D47516633
Pull Request resolved: https://github.com/pytorch/pytorch/pull/105375
Approved by: https://github.com/jansel