perf/x86/intel: Fix broken fixed event constraints extension

Unnecessary multiplexing is triggered when running an "instructions"
event on an MTL.

perf stat -e cpu_core/instructions/,cpu_core/instructions/ -a sleep 1

 Performance counter stats for 'system wide':

       115,489,000      cpu_core/instructions/                (50.02%)
       127,433,777      cpu_core/instructions/                (49.98%)

       1.002294504 seconds time elapsed

Linux architectural perf events, e.g., cycles and instructions, usually
have dedicated fixed counters. These events also have equivalent events
which can be used in the general-purpose counters. The counters are
precious. In the intel_pmu_check_event_constraints(), perf check/extend
the event constraints of these events. So these events can utilize both
fixed counters and general-purpose counters.

The following cleanup commit:

  97588df8 ("perf/x86/intel: Add common intel_pmu_init_hybrid()")

forgot adding the intel_pmu_check_event_constraints() into update_pmu_cap().
The architectural perf events cannot utilize the general-purpose counters.

The code to check and update the counters, event constraints and
extra_regs is the same among hybrid systems. Move
intel_pmu_check_hybrid_pmus() to init_hybrid_pmu(), and
emove the duplicate check in update_pmu_cap().

Fixes: 97588df8 ("perf/x86/intel: Add common intel_pmu_init_hybrid()")
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20230911135128.2322833-1-kan.liang@linux.intel.com