1. Try obtaining crystal clock frequency from CPUID_TIME_STAMP_COUNTER (15h) leaf from ECX.
2. On failure, take predefined values for known processor models.
3. On failure, check for CPUID_PROCESSOR_FREQUENCY availability and divide by the TSC ratio.
4. On failure, assume 24 MHz.
Added info about the highest supported leaf (OC_CPU_INFO::MaxId).
Fixed a few places which were incorrectly comparing standard CPUID leaves against the maximum extended function leaf (OC_CPU_INFO::MaxExtId).
Add a warning in case the TSC- and ART-derived frequency calculations are wildly different.
Correct CPU frequency calculation. TSC Adjust is only for converting timestamps between ART and TSC, not calculating the CPU frequency.
Only apply the 25Mhz ART fallback if the processor is a Xeon Scalable, determined via the brand string until we can find a better method.
Previously, OcCpuLib assumed all CPUs had the same core crystal clock frequency for their ART (24 Mhz). Xeon Scalable and Intel Atoms with Goldmont architecture have different frequencies (25 Mhz and 19.2 Mhz respectively). If the CPU supports reporting its frequency, use that. Otherwise, fallback to a value based on its model and family.
Cleaned up the logic for how we compute our CPU and FSB frequencies. For CPUs that support ART, use that instead of the TSC frequency.
Added documentation and renamed `OC_CPU_INFO::TSCFrequency` to `CPUFrequencyFromTSC` and added `CPUFrequencyFromART` for clarity. These are intermediate values used to compute `CPUFrequency`. `ARTFrequency` is now correctly set to the core crystal clock frequency rather than `CPUFrequencyFromART`, which it was previously.
Lastly, adjusted the `CPUFrequencyFromART` calculation to include the TSC offset in `MSR_IA32_TSC_ADJUST` if present. I'm not aware of which CPUs have a non-zero offset but it's part of Intel's calculation for computing the CPU frequency based on the ART.
