- no significant perf change with this or preceding cache mode changes, but
  including for completeness

- mislabeled ofi/verbs results as ofi/psm2

- not included in nightly tests yet, requires use of evaluation nodes
  that aren't in production
- tests run cleanly but bandwidth is slow

- experimental, just for benchmarking

- current origin/master seems unstable for psm2 in libfabric

- add bw tests
- switch cci tests to busy-spin only

- set env variable for GNI provider to work alongside MPI
- add bandwidth tests

(need to make similar updates for static build version)

- timing now performed by client to get effective rate including rpc
  req to start each stream and rpc ack to complete it
- intent is to more accurately measure transfer rate instead of just
  injection rate

- put comments in lookup3.c
- replace uthash.h with newer version

- needs testing

- this change is in origin/master now