Lasagne backend doesn't iterate over all data when you set batch_size

[jmgore75]

In pull request #157 I describe how when you use a batch size, the lasagne backend does not train with all of the data, only blocks of that batch size (leaving out the bit at the end). Also, if batch_size is greater than the total data size, it won't iterate over any data at all and will die with a zero division error.

[alexjc]

Yep, thanks for pointing it out. Just needs one test to make sure we catch such problems in the future.

[alexjc]

If you want me to merge this from your PR, please include the three tests from [7831214]. Thanks again!