Decouple the computational batch size and minibatch size by accumulating gradients

[longjon]

After #1615, so that this code already supports deconv layer. (The actual diff is just +37/-40 lines.)

This PRs the gradient accumulation branch living at https://github.com/shelhamer/caffe/tree/accum-grad. I took a lighter approach here than the one there: parameter gradients are always accumulated, there is no other option. The gradient checker is made correct by zero-initing parameter diffs.

Issues:

• This changes the behavior of Backward. External code that used Backward is likely to break, if there is any.
• I think this breaks solvers other than SGDSolver, but haven't thought carefully about that yet.

[jeffdonahue]

Have we thought about how to handle the case when we're sharing parameters but using different learning rates? I would be okay with simply disallowing that case since it would probably be a pretty weird thing to do. Otherwise the only other way I can think to handle it is pretty messy -- we could have a a special case where, e.g. if blobs_lr is 2 in one layer but 1 in all others, the Net could prescale (by a factor of 2) the top_diff for the layer with blobs_lr 2 by 2... Actually, even that wouldn't work if the layer has other shared param blobs that don't also have the same relative LR...

[shelhamer]

Always accumulating is simple and good, but let's review the weight sharing and solvers issues before merging.

[shelhamer]

Replaced by #1977.