RDMA/mlx5: Remove redundant work in struct mlx5_cache_ent

delayed_cache_work_func() and the cache_work_func() are both wrappers of
__cache_work_func(). Instead of having a special not delayed work, use the
delayed work with delay = 0.

Link: https://lore.kernel.org/r/18b6ae205e75f087aa4a2a05c81ea8b66d8d88dc.1644947594.git.leonro@nvidia.com


Signed-off-by: Aharon Landau <aharonl@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>