Not enabled by default as for some case it is 1.17x the original time(slowdown) but in other case it is .07x the original time(13 times faster!)

replaced the theano.gof.utils.get_theano_flags fct by a class theano.gof.utils.[TheanoConfig,config] that is a variante of ConfigParse.

if the env variable THEANO_FLAGS have the token local_elemwise_fusion we enable the fusion. Each token is separated by comma.

function is to flatten the output, but I don't think that is generally wanted or
useful.  The old tests use it, but I don't think people should be left wondering
whether they should use DownsampleFactorMax or max_pool.

bigger job.  See TRAC ticket on subject.