Switch self profile to use HW counters instead of walltime (attempt 2)

[Kobzol]

This is a second attempt to land https://github.com/rust-lang/rustc-perf/pull/1647, which was later reverted in https://github.com/rust-lang/rustc-perf/pull/1700.

I tried to manually SSH into the collector server and run the whole benchmark suite with --self-profile using icounts. It went through. I'd like to try this on the live server to see what happens. If it fails again, we'll know more to see where we should investigate the failure.