perf macro

[vladimir-ea]

adds a #[perf] macro that can be used like the #[timed] one. This one is gated behind a feature flag perf.

use it like:

    #[timed]
    #[perf]
    pub fn decompress() { 

or:

    #[timed]
    #[perf(sample = 100)]
    pub fn decompress() { 

to record every 100 times the function is called.

outputs: instruction count, cpu cycles, instructions per cycle (ipc), branch misses, cpu cache misses on surfer 'perf' tab: