Change tracer consistency system test from hobart_nag to hobart_intel

This started failing with nag, for reasons I couldn't understand (see
#789). Rather than trying to dig
into this hard-to-explain issue, I'm just changing the test to intel,
where it passes. (It also passes using gnu on my mac.) Note that the
memory leak also goes away with intel.

We don't have many other hobart_intel tests, but it seems fine to me to
start introducing more, in order to avoid increasing our testing time on
cheyenne.

cime_config/testdefs/testlist_clm.xml

@@ -1668,17 +1668,10 @@
   <test name="SMS_D_Ld10" grid="f10_f10_musgs" compset="I2000Clm50BgcCropGs" testmods="clm/tracer_consistency">
     <machines>
 
-      <machine name="hobart" compiler="nag" category="aux_clm">
+      <machine name="hobart" compiler="intel" category="aux_clm">
         <options>
           <option name="wallclock">0:30:00</option>
           <option name="comment">Include a tracer consistency check in debug mode.</option>
-          <!-- There is a memory growth of about 12% over this 10-day
-               test, just with nag. This may be a real issue, or it may
-               be a compiler issue. For now, since there's a chance it's
-               just a compiler issue, we're allowing this level of
-               memory leak. See
-               https://github.com/ESCOMP/ctsm/issues/763 for details. -->
-          <option name="memleak_tolerance">0.3</option>
         </options>
       </machine>