intel: base shader cache key on intel_device_info fields, stub gpu from serialized json
This MR solves 2 problems that prevent implementation of an cloud-based multi-platform shader cache:
- Current use of PCI ID as the key to differentiate platforms leads to highly redundant cache entries
- Current mechanism to stub the gpu based on PCI ID is only used to estimate shader costs, and is not accurate.
These issues are solved by basing the shader cache key on the intel_device_info fields that are accessed under src/intel/compiler, and by populating those fields from serialized json when running with a stubbed gpu.
closes: #6746
Edited by Mark Janes