MSR reads from parked cores wake the cores unnecessarily

On LNL, with P-cores offline via cgroup cpuset (I have a local patch that makes `nohz_full=0-3` work), we're still periodically reading MSRs from each. The P-core MSR read does an IPI, which wakes the core that we're trying to keep asleep in the first place. Can we stop polling MSRs for CPUs that are supposed to be quiet?

(Also, as a separate matter, we should mask the thermal interrupt via LVT, right? Otherwise we get useless HFI interrupts on these cores that are supposed to be asleep.)