Exceptions not thrown/SIGABRT when concatenating vectors of different batch sizes

Example for not throwing exceptions:
```python
import dynet as dy

x1 = dy.zeros((4,), batch_size=3)
x2 = dy.zeros((4,), batch_size=2)
x = dy.concatenate([x1, x2])
print(x.dim()) # prints ((8,), 3), but should throw an exception
```

And, in a part of a larger program, I tried concatenating expressions of batch sizes 10, 10, 2, and 1. The first three are complicated expressions involving LSTMs, and the last one was a `dy.lookup`. In this case, the program aborted given the following message:
```
Assertion failed: (dimensions_match(m_leftImpl.dimensions(), m_rightImpl.dimensions())), function evalSubExprsIfNeeded, file /usr/local/Cellar/eigen/HEAD-1ff051b7a3ae/include/eigen3/unsupported/Eigen/CXX11/src/Tensor/TensorAssign.h, line 122.
```
I'll try and see if I could come up with a minimal example.

Start with the dy.concatenate call in the Python example and trace how mismatched batch sizes reach the Eigen assertion in evalSubExprsIfNeeded. Compare the handling of batch sizes 3 and 2, and of 10, 10, 2, and 1; done means mismatches are rejected with an exception instead of silently producing a result or aborting.

Written by the indexing model from the issue text.