Finishes closing of #539.

Closes #539.

Closes #538 .

a < b < c was failing silently by generating the graph "lt(b,c)".  We now forbid
"boolean" comparison of the return value of __lt__, __le__, __gt__, __ge__.
We cannot forbid "boolean" comparison of TensorVariables in general, since the
syntax "if a:" is often used instead of "if a is not None:"

replace the theano flags warn.old_bug_default by warn.ignore_bug_before and doc it. Also update the installation instruction to have less error warning generated during the test of the installation.

make the Convolution use all needed compilation option when it use gemm. Generate them only when we generate call to gemm.

new theano flags: `warn.old_bug_default`. When set to false disable all warning about old theano bug.

Fixes http://trac-hg.assembla.com/theano/ticket/155

numpy.ndarray overrides __eq__ to do elemwise equality checks, thus yielding an ndarray, which cannot be cast as a bool. This doesn't play nicely with list.index; numpy throws a ValueError.

numpy.ndarray overrides __eq__ to do elemwise equality checks, thus yielding an ndarray, which cannot be cast as a bool. This doesn't play nicely with list.index; numpy throws a ValueError.

Closes ticket #518.

make the split of too huge GpuElemwise check about the number of dimensions of the inputs. test this too.

Added fix suggested by Pascal for the tests when dealing with the exception thrown by Subtensor (e_invalid)

added test for same bug

more helpful.