Fix how multipart samples (IBS) are stored:
- Clear and count the whole 8-byte header (two slots on 32-bit).
- Count the payload when the callchain depth is 1.
- Do not reserve an extra slot for the user callchain.
- Use the real PC, not the header, for profil(2).
- Make sure a sample slot can hold the largest payload.
Add <sys/pmclog.h> helpers to find the callchain in a multipart record.