venus: optimize vn_relax strategy for different scenarios
What does this MR do and why?
venus: optimize vn_relax strategy for different scenarios
Summary:
- commits 1~2 allow venus ring to idle out earlier without non-trivial penalty for wakeup
- commits 3~4 updates query pool result polling to use vn_relax
- this improves gfxbench manhattan scores with A-on-V by ~15% on some AMD platforms
- commits 5~6 deprecates unused perf env vars and decorate no_cmd_batching with compiler hints
- last commit adds a more relaxed polling strategy for all other waits except ring seqno waiting