fix(ebpf): improve pcindex binary search speed

[korniltsev]

PCIndex.FindIndex is not super hot, it is only 3%. However the change is not hard so I do it anyway. "It ain't much, but it's honest work"

In this PR I've inlined sort.Search in PCIndex.FindIndex.

goos: linux
goarch: amd64
pkg: github.com/grafana/pyroscope/ebpf/symtab/gosym
cpu: AMD Ryzen 9 5950X 16-Core Processor            
             │   old.txt    │               new.txt               │
             │    sec/op    │   sec/op     vs base                │
BinSearch-32   3.354m ± 11%   1.716m ± 1%  -48.83% (p=0.000 n=10)

[kolesnikovae]

Yeah, sort.Search sometimes makes me sad. I wonder if slices.BinarySarch would be of use here – it was released as part of 1.21.0, for versions before there is x/exp/slices package

[korniltsev]

yeah slices do work, thanks

goos: linux
goarch: amd64
pkg: github.com/grafana/pyroscope/ebpf/symtab/gosym
cpu: AMD Ryzen 9 5950X 16-Core Processor            
             │   old.txt    │           new_slices.txt            │
             │    sec/op    │   sec/op     vs base                │
BinSearch-32   3.428m ± 14%   1.765m ± 2%  -48.50% (p=0.000 n=10)

I also realized that there was small inconsistency for different symbols with same addresses (for example function aliases). For example if function is at 0xff0 and it is 0x10 long and it has two names foo bar and I search for 0xff0 the pcindex would return 'foo' and if search for 0xff1 return bar . Maybe I should just remove the the unused dupes, but in a separate PR.