Always use RMA when UCX is selected

[wangvsa]

To support the UCX data transfer layer, we currently offer both regular UCX and UCX with RMA.
I suggest we remove the compile-time option DYAD_ENABLE_UCX_DATA_RMA and always use RMA when UCX is selected. Since UCX+RMA consistently outperforms UCX without RMA, this change makes sense.

We’ve already completed the experiments and included the performance comparison in the paper, so there’s no longer a need to maintain two separate UCX code paths. This would make the UCX dtl code much easier to understand and maintain.

[wangvsa]

@JaeseungYeom is this okay to you? if so, I'll clean the code.