osd: fix UBSan shift overflow in pg_pool_t::calc_pg_masks

UBSan reports: shift exponent 32 is too large for 32-bit type 'int'

When pg_num or pgp_num is 0, pg_num-1 wraps to UINT_MAX (unsigned),
cbits() returns 32, and (1 << 32) on 32-bit int is undefined behavior.

Use 1ULL for the shift so the expression is well-defined; for cbits()
returning 32, (1ULL << 32) - 1 equals UINT_MAX.

Signed-off-by: Alexander Indenbaum <aindenba@redhat.com>