nir/serialize: Make more space for intrinsic_op allowing 1024 ops

We are close to the limit of 512 intrinsics, make more space to be able to support up to 1024 intrinsics.

Take one bit from packed_const_indices, they shouldn't suffer in a common case.

Motivated by the need to introduce _ir3 versions of global_atomic_X intrinsics in order to have 32b vec2 global address after lowering global 64b address (!8717 (comment 1106358)). Adding them pushes us over 512 intrinsics limit.

Other suggestions are welcomed.