Bug in MRNN, the model should iterate on each feature

[WenjieDu]

1. System Info

All PyPOTS versions.

2. Information

• The official example scripts
• My own created scripts

3. Reproduction

In the official implementation, MRNN iterates on each feature https://github.com/jsyoon0823/MRNN/blob/77227d49b99053cbfe4380255a247e819950e074/mrnn.py#L292-L294

4. Expected behavior

But in the PyPOTS' version, MRNN runs on all features.

[WenjieDu]

Didn't find any performance improvement on real-world datasets after fixing this issue. But the running speed is greatly slowed down because it needs to iterate *n_features times.