turnip/perfetto: Perfetto support for KGSL
On KGSL the raw GPU time could be queried via kgsl_cmdbatch_profiling_buffer
on submission. I refactored a bit our perfetto code to accept the GPU time in a more straightforward way to be usable for KGSL.
KGSL also returns CPU time together with GPU's, but I don't use it at the moment.
Wild Life benchmark:
(There are sometimes artifacts but they also happen with msm)