Concatenate batch fails if vectors are of sizes (dim,) and (dim, 1)

Hi! The title sums it up. If you try to concatenate d-dimensional vectors but some vectors' shape is (d,) while others is (d, 1), dynet.concatenate_batch fails. However, if you do dynet.concatenate_cols and then a reshape, everything works fine.

Start by reproducing the failure with dynet.concatenate_batch using vectors shaped (dim,) and (dim, 1), then compare it with dynet.concatenate_cols followed by reshape. Trace the concatenate_batch entry point and identify the expected behavior for mixed vector shapes. Done means concatenate_batch accepts both shapes consistently without requiring the workaround.

Written by the indexing model from the issue text.