Test more case of this Op.

is local and already dead (in which case we steal the lock immediatly).
Otherwise the old comportement is kept (steal the lock after 120 seconds
of inactivity).

The code is made to deal with old locks it could encountered by treating
those locks as coming from other hosts so there should not be any functional
changes.

modified CudaNdarray_Reshape to accept int as a shape object. Test this case and test that the GpuFlatten work correctly as it use this code.

allow the lib.amdlibm flag to be on when using nvcc. We disable it for nvcc as they don't work correctly together in device code.

This will remove most of the error message that was changed today for a warning.

fix a bug where we pushdown the max to the input of softmax, softplus, tensor.exp, tensor.log, tensor.tanh, sigmoid, softmax_with_bias.

Add test for this bug and print a warning by default if we encounter the case where the bug was causing problem. The warning can be disabled by the Theano flag warn.argmax_pushdown_bug.

Add test for CAReduce{maximum,mul} and modified CAReduce to handle correctly maximum.