Fix stale goroutine label, validate UTF-8 on label strings

[gnurizen]

Summary

Backport of open-telemetry/opentelemetry-ebpf-profiler#1454 to the parca-dev fork.

This regressed in 40dd9d9 (open-telemetry#1091), which removed the explicit zeroing of trace->custom_labels.labels on the assumption that the slots were only read after a successful write that fully populated them. The slots are populated via bpf_probe_read_user, which only writes the bytes it reads, so a key/value shorter than one previously written to the same per-CPU slot inherits trailing bytes from the prior trace and produces a corrupted label.

• eBPF side: write a single NUL after each bpf_probe_read_user. Userspace stops at the first NUL, so one terminator is sufficient. The fork's CustomLabel buffer is already sized MAX_LEN + 1 to make room for the terminator, so the read clamp stays at MIN(len, MAX_LEN) (no -1).

• Go side: validate label keys and values as UTF-8, which OTLP/pprof require. Strictness is deliberately asymmetric:

  • Keys are strict. Any invalid byte (including a single split-rune continuation at the end) drops the whole label. A corrupted key would silently group unrelated samples under a garbage name.
  • Values are lenient. Fixed-width eBPF buffers can clip a multi-byte rune in half; we salvage the longest valid UTF-8 prefix rather than discard the label, so a clipped request_id or customer_name arrives intact up to the broken rune.

• Two new metrics (IDGoLabelsDroppedInvalidName = 305, IDGoLabelsDroppedInvalidValue = 306) count labels dropped for each reason. Renumbered from upstream's 293/294 because the fork already uses those IDs for LuaJIT counters.

• comm is left as best-effort: kernel-supplied, almost always ASCII, no useful fallback if it isn't.

• Also disabled 5.4 kernels in testing because they have issues with this change.