fix(ebpf): fix issue when a cls arg is a cell

[korniltsev]

Fixes "invalid utf string" issue in our dev cluster.

The issue was caused by ebpf profiler type confusing - trying to use PyCellObject as PyTypeObject

In addition fixes issue when first argument name is incorrectly detected by zeroing the arg name buffer. (previously if could contain garbage from previous frame)

[simonswine]

LGTM

Couldn't find a regression, with a quick happy path check. And I also spotted this for the first time 🙂

pyroscope/ebpf/bpf/pyperf.bpf.c

Lines 295 to 300 in a189d57

	// Figure out if we want to parse class name, basically checking the name of
	// the first argument,
	// ((PyTupleObject*)$frame->f_code->co_varnames)->ob_item[0]
	// If it's 'self', we get the type and it's name, if it's cls, we just get
	// the name. This is not perfect but there is no better way to figure this
	// out from the code object.

[korniltsev]

I preproduced the issue here https://github.com/pyroscope-io/pyroscope-ebpf-testdata/blob/fc63fce7e3e3995d305cdfab0efc239e5632c164/rideshare-flask-no-pip/lib/server.py#L28-L43

[korniltsev]

And I also spotted this for the first time 🙂

Yesterday I found another issue with the way the code detects class name.

f_code->co_varnames may contain same name multiple times. It may look like ('self', 'foo', 'bar', 'self') and f_localsplus[0] == NULL and f_localsplus[3] == 0x7f....