The old form (PatternSub) had the problem that it wasn't calling broadcast_like
on the return value.  It would have been wrong to hack a broadcast_like into
gof.opt, so I rewrote these two optimizations as a function.

PatternSub is a really cool idea, in the future it might be worth writing a
TensorPatternSub that deals with tensor-specific issues, like casting,
dimshuffles, and rebroadcasts.

Review/tests for ExtractDiag and trace

added a hint to the BadOptimization exception

Fix div by zero in tensor grad

Correctly do version comparison: fix #136

misc cleanup and a bugfix in linalg ops

added sparse dot

Op doc merge

Conflicts:
	doc/cifarSC2011/extending_theano.txt
	doc/extending/op.txt

moved import cuda where I need it.

function

Op documentation

More explicit error message

Conflicts:
	theano/sparse/basic.py

Makes it more obvious to users what may be their mistake when they get a
TypeError after trying to index a shared variable.