add per cpu ringbuffer

[dshulyak]

one serious problem with it, i have to lower memory per ringbuffer, otherwise paged memory for profiler grows unreasonably.
if there is a burst of events on single cpu i will be missing them, when i wouldn't miss them if i was using single buffer.

contention on ringbuffer also doesn't appear to be an issue, maybe i need better tests for that, but overall i don't see a reason to merge this

2024-07-28T13:30:46.311491Z  INFO past::bpf_profile: BPF PROFILE:
program         | duration s      | runtime s       | busy %          | runtime cnt     | collected cnt   | collected latency ns
profile         | 10.4589         | 0.0037          | 0.04            | 24840           | 0               | inf            
exec            | 10.4589         | 0.0000          | 0.00            | 2               | 1               | 31971.00       
switch          | 10.4589         | 0.0259          | 0.25            | 132442          | 36              | 719661.64      
rss             | 10.4589         | 0.0599          | 0.57            | 2862210         | 87              | 688622.77