linux: performance monitoring API

[vavrusa]

This PR implements perf_event_open(2) and a higher level API. It also creates a perf reader metatype as a high-level interface modelled after perf(1) and libbpf perf reader interface.

• Added constants and FFI types for creating a struct perf_attr
• Added metatype for struct perf_attr for easy filling (strflag conversions)
• Added perf event reader, support both read(2) and mmap(2) interfaces
• Added support for performance counters (software, hardware, hardware caches, ...)
• Added function to attach to tracepoints (requires debugfs)
• Function to attach/detach kprobes and uprobes
• Parser for mmap(2) sampling interface, it can read events off ringbuffer with standard ipairs()

I'm not adding perf event ABI dissectors here because it's (a) really complicated and unstable and (b) events have variable sizes, so generic parser may be cost-prohibitive. Caller gets standard event header that contains event type, size and raw contents. I'll probably work on parsing code for more complicated event types, but as a separate PR.

[vavrusa]

Is this something you're considering or not at all? It's groundworks, but I'd really like to merge friendlier interface[1] when the foundations are done.

[1] i.e. statically compiled event readers have to make wild guesses (section "Usage examples") about how probe ABI looks like, with LJ we can parse the format and turn it into ctype on runtime, that'd be really cool

[justincormack]

Sorry, I forgot to look at it again, looks like a good start, will experiment with it a bit too.