reporter: emit CUDA PC-sample frames (cuda-pc) for per-function symbo…

…lization (#3186)

Add the CUDAPCFrame case to both location-writer paths (appendLocationV2
and the V1 record path): one mapping per cubin (build_id = cubin CRC
FileID), the kernel's mangled name as the system name of a placeholder
line (line 0, no function name), and the "cuda-pc" frame type. The
backend resolves the source line per function off the kernel name.

The stall reason and SASS instruction arrive as the cuda_stall_reason /
cuda_sass_instruction sample labels (trace.CustomLabels, handled
generically), independent of the frame.

Bumps the ebpf-profiler fork to pick up libpf.CUDAPCFrame and the
single-frame GPU PC-sample emission (parca-dev/opentelemetry-ebpf-profiler#300).
Pinned to a main commit pending a tagged release.

Bump fork (#3171)

fix: avoid nil dereference in symbol uploader when temp file creation…

… fails (#3163)

`attemptUpload` creates a temporary file before extracting debuginfo. If
`os.Create` fails, the current error path calls `os.Remove(f.Name())`,
but `f`
is `nil` in that case, so `f.Name()` panics and crashes the agent.

I ran into this on a NixOS host after parca-agent panicked in the symbol
uploader while processing a wrapper binary under `/run/wrappers/...`.
The
stack trace pointed back to the `os.Create` error path in
`reporter/parca_uploader.go`.

This change removes the `os.Remove(f.Name())` call from the `os.Create`
error
branch. When file creation fails, there is no file to remove, and the
function
now returns the original `create file` error instead of panicking.

Update fork to pick up batch kernel timing probe changes (#3151)

Update Go to v1.24.13 (#3142)

## Summary
- Update Go version from 1.24.7 to 1.24.13

bump fork

bump fork

Update oomprof and otel for oomprof gobuild id fixes (#3117)

Bump to latest otel build (#3115)

Bump opentelemetry-ebpf-profiler to bd15b7e3a862 (#3109)

Picks up:
- bd15b7e Fix custom labels for node v25 (#165)
- 8f82833 Bump key size to 25 (#162)