AMDFamily0F: compare the K8 model against 0x40, not decimal 40

The core select encoding changed with NPT at model 0x40, but the check
was written as decimal 40. Models 0x28-0x3F, Manchester 0x2B among them,
therefore took the post-NPT mapping and asked for the wrong core, so
ioctl_get_thermtrip could return the other core's temperature or select
a core that isn't there.