Fix DF-4251: mprsas_SSU_to_SATA_devices leaks mpr_lock on two error paths.

The lock is taken at the top (mpr_sas_lsi.c:1479); the xpt_alloc_ccb()
failure return (:1501) and the xpt_create_path() failure return (:1510)
both bypass the only mpr_unlock() after the target loop (:1541). Runs
from mprsas_ir_shutdown() at system shutdown: after either error return
the adapter lock is permanently owned by a dead thread; every later
mpr_lock user (mpr_intr, the periodic callout, cam teardown/detach)
blocks forever and shutdown wedges.

Verified live (rig, baseline kernel #0): the REAL production
mprsas_SSU_to_SATA_devices, dispatched under exactly the production
xpt_create_path failure condition (unregistered bus id ->
CAM_PATH_INVALID), returned with lockstatus()==LK_EXCLUSIVE and a kernel
thread then blocked on mpr_lock indefinitely until the rig released it.
With this fix the same dispatch returns with the lock released
(lockstatus()==0).

--- a/sys/dev/raid/mpr/mpr_sas_lsi.c
+++ b/sys/dev/raid/mpr/mpr_sas_lsi.c
@@ -1498,7 +1506,7 @@
 			if (ccb == NULL) {
 				mpr_dprint(sc, MPR_FAULT, "Unable to alloc CCB "
 				    "to stop unit.\n");
-				return;
+				goto out_unlock;
 			}
 
 			if (xpt_create_path(&ccb->ccb_h.path, xpt_periph,
@@ -1507,7 +1515,7 @@
 				mpr_dprint(sc, MPR_ERROR, "Unable to create "
 				    "path to stop unit.\n");
 				xpt_free_ccb(&ccb->ccb_h);
-				return;
+				goto out_unlock;
 			}
 			xpt_path_string(ccb->ccb_h.path, path_str,
 			    sizeof(path_str));
@@ -1538,6 +1546,7 @@
 		}
 	}
 
+out_unlock:
 	mpr_unlock(sc);
 
 	/*