perfetto: Refactoring common data source code

I needed to borrow Intel's app_event stuff for getting command stream annotation in tu, and got frustrated at all the copy and paste in GPU command stream datasources. This isn't any big code removal, but hopefully the comments help people make sense of the actually driver-specific stuff, even if it means some ugly templated static method calls to get at shared code. And, now that I've fought the c++ compiler a bit, hopefully others will be able to write new shared code, too.