fabtests/efa: Add multi-QP support for efa_rma_bw

Add support for multiple endpoints (-q / --num-eps) with per-EP flow
control and FI_MORE doorbell batching. Each EP independently tracks its
posted/completed counts via an efa_rma_bw_ctx embedded in the operation
context, with pool slots partitioned per-EP to prevent cross-EP reuse.
The benchmark loop follows the perftest pattern: iterating all EPs each
pass to avoid starvation, with per-EP window checks gating both posting
and FI_MORE. The writedata rx path uses fi_recvmsg with per-EP tracking
when FI_RX_CQ_DATA is set.

Signed-off-by: Shi Jin <sjina@amazon.com>

Author: shijin-aws