Bug in `em1s_rotate`

This problem was found by the Trento group. 

The rotated `em1s` from yambopy seems to be inconsistent with that of a native yambo nosym calculation, at least for the case of no spatial inversion (meaning trev is used for the `em1s` rotation).

In particular, this is the inconsistency:
```
X_mat_yambopy(iq,iG1,iG2) = X_mat_yambo(iq,iG2,iG1)
``` 
which leads to different BSE absorption spectra.