Scan crash fix

Use half_nbits floats instead of nbits. fixes gh-1510

Use compat.python2x.all instead of numpy.all on generator

Sparse updates

Otherwise, the generator evaluates to "True", because numpy.all does not
actually iterate through it.

Also add test case to check that m[0, idx_list] triggers full
advanced indexing, not "take".

This help test those cases.

Conflicts:
	theano/sparse/tests/test_basic.py

Doc how to make an sparse op

Use [:mintap] instead of [:-n_tap_done], to make it use the same was as other place to help subtensor opt.

FIX: Example code from tutorial page fails with conversion error

I updated the test to check that. Other place where we talk about remove0 was telling that it just remove 0.

Add documentation about the fct csm_data, csm_indices, csm_indptr, csm_shape, csm_properties and theano.sparse.tests.test_basic.sparse_random_inputs.

This help test those cases.

fix gh-1493 tutorial example that use undefined variable.

Fix exception/segfault newly introduced in gpujoin.

use absolute reference for abs_

For some reason this was failing before

http://deeplearning.net/software/theano/tutorial/examples.html fails with
conversion error, this change fix the problem.

SymPyCCode op

Change grad function name to "grad", not "prime"

Change y**2 test to y**3

Backport to python 2.4