u_trace,vk/wsi,tu,anv: Plumb proper frame delimiters

What does this MR do and why?

Previously u_trace had to think that every submission is a separate frame, which makes it hard to analyze results at frame granularity. E.g. if I want to compare tracepoint times between same frames. There was already a frame counter in WSI, it just ran only when trace_mode was set (which is changed here).

Why not use eof param in u_trace_context_process? It works for gallium, but in VK we chose to call u_trace_context_process at submission time where we don't know if it's a new frame or not.

@llandwerlin could you test that u_trace + anv prints frames correctly with my change?