More cases are detected now.

zeros.

new optimization for dot products.

This flag is going to be used by optimizations. For example dot(zeros, X)
should be optimized out even if zeros is a constant.

Always update the following properties of current mode, even if current
function does not have profiling statistics: call_time, fn_time,
optimizer_time, linker_time. Re-add these attributes in ProfileMode
(they exist in Mode).

This basic timing functionality is used in particular for timing the deep
learning tutorials.

Also fix error that was causing compilation cache to fill up.

We should pass the Mode objects instead.

We should use, for instance, "linker=cvm_nogc", not "mode=CVM_NOGC".

Also get rid of all those debug(), info(), warning() and error() functions
defined everywhere.

In the past it was warned when we load theano.sandbox.cuda.

Consecutive calls of the the join op ( during the optimization phase) get
the axis as a constant. What used to happen was that in such cases Theano
would not use that constant to define its broadcastable pattern.