mercury: HG_Core_context_destroy(): HG core handles must be freed before destroying context (256 remaining)

Describe the bug Recent mercury changes now result in the “margo-p2p-bw” test (and lots of others) spewing hundreds of lines about handles that are not freed.

This did not happen a few weeks ago. Do these errors seem like the kind of thing that would show up after some of the recent changes in mercury?

 # HG_Core_context_destroy(): HG core handles must be freed before destroying context (256 remaining)
# HG -- Error -- /tmp/cscstdm/build-stage/spack-stage-mercury-master-aa3qembnn5skb7drjreqrxnl4ryocdgm/spack-src/src/mercury_core.c:3720
 # HG_Core_context_destroy(): HG core handle at address 0xe12e00 was not destroyed
# HG -- Error -- /tmp/cscstdm/build-stage/spack-stage-mercury-master-aa3qembnn5skb7drjreqrxnl4ryocdgm/spack-src/src/mercury_core.c:3720
 # HG_Core_context_destroy(): HG core handle at address 0xe12690 was not destroyed
...
# HG -- Error -- /tmp/cscstdm/build-stage/spack-stage-mercury-master-aa3qembnn5skb7drjreqrxnl4ryocdgm/spack-src/src/mercury.c:1105
 # HG_Context_destroy(): Could not destroy HG core context (HG_BUSY)
# HG -- Error -- /tmp/cscstdm/build-stage/spack-stage-mercury-master-aa3qembnn5skb7drjreqrxnl4ryocdgm/spack-src/src/mercury_core.c:1172
 # hg_core_finalize(): HG contexts must be destroyed before finalizing HG (1 remaining)
# HG -- Error -- /tmp/cscstdm/build-stage/spack-stage-mercury-master-aa3qembnn5skb7drjreqrxnl4ryocdgm/spack-src/src/mercury_core.c:3493
 # HG_Core_finalize(): Cannot finalize HG core layer
# HG -- Error -- /tmp/cscstdm/build-stage/spack-stage-mercury-master-aa3qembnn5skb7drjreqrxnl4ryocdgm/spack-src/src/mercury.c:1001
 # HG_Finalize(): Could not finalize HG core class

The tests appear to complete without error, but these messages completely swamp the output we’re looking for.