lib/nvme: Fail ctrlr reconnection attempt early if ctrlr is failed

During a reset of a fabrics controller, we disconnect the controller and
then we attempt to reconnect.
The disconnection will unset the `is_failed` flag in the controller and
the adminq is disconnected.
If something bad happens in the meantime, the controller may be marked
as `is_failed` again.
Then when the reconnection start, we reset the adminq's state to
CONNECTING and we start polling on it until we get a response to our
connection request. But since the controller is marked as is_failed and
the adminq's state is CONNECTING, the polling will not do anything (cf
`spdk_nvme_qpair_process_completions()`).
Moreover, the controller is in a WAIT_FOR_CONNECT_ADMINQ state with an
infinite timeout. So the controller may be blocked forever.
Let's try to prevent this situation by checking the `is_failed` flag
before attempting a reconnection.

Change-Id: Id83ff161e0b389fa2e266468006f619ad6bc65c1
Signed-off-by: Alex Michon <amichon@kalrayinc.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/24649


Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Jim Harris <jim.harris@samsung.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Shuhei Matsumoto <smatsumoto@nvidia.com>