[ofi 2.0] Make fabtests/benchmarks able to use single CQ per EP

[shijin-aws]

Currently, fabtests still use separate tx and rx cq per EP for benchmarks.

For ofi 2.0, we propose to restrict binding 1 CQ per EP. The current fabtests should at least be configured to allow binding 1 CQ per EP for benchmarks.

[j-xiong]

Unrelated to the CQ change for 2.0. More in line with current usage. Could be complicated to distinguish send/recv completions.