Add userland glue for the new in-kernel PMU group/multiplex machinery:
pmc_allocate_group() wrapper around pmc_allocate that or-s
PMC_F_GROUP_DEFER into the flags.
pmc_group_create() allocate a kernel group id.
pmc_group_add() bind a pmcid (allocated with
pmc_allocate_group) to a group, optionally as the leader.
pmc_group_commit() freeze membership; the kernel scheduler is
free to place the group on hardware afterwards.
pmc_group_release() tear down a group and detach its members.
In addition, pmc_read() returns the time-scaled count when the kernel
sets PMC_F_SCALED on a read, and the new pmc_read_pair() returns the
raw count together with time_enabled / time_running for callers that
want to do their own scaling.
libpmc is not symbol-versioned, so the new entry points are exported
automatically; no map file needs to change.
Sponsored by: AMD
Signed-off-by: Raghavendra K T <raghavendra.kt@amd.com>