Add OpFromGraph in the doc.

doc/library/compile/opfromgraph.txt

+.. _opfromgraph:
+
+===========
+OpFromGraph
+===========
+
+This page descripbe :class:`theano.OpFromGraph
+<theano.compile.builders.OpFromGraph>`. an Op that allow to
+encapsulate a Theano graph in an op.
+
+This can be used to encapsulate some functionality in one block. It is
+useful to scale Theano compilation for regular bigger graph when we
+reuse that encapsulated fonctionality with different inputs many
+times. Due to this encapsulation, it can make Theano compilation phase
+faster for graph with many nodes.
+
+Using this for small graph isn't recommanded as it disable
+optimization between what is inside the encapsulation and outside it.
+
+.. note:
+
+    This wasn't used widely up to now. If you have any
+    questions/comments don't contact us on the mailing list.
+
+
+
+.. autoclass:: theano.compile.builders.OpFromGraph

doc/library/compile/opfromgraph_example.py

+import theano
+from theano import tensor, OpFromGraph
+
+
+x, y, z = tensor.vectors("xyz")
+
+out1 = x + y
+out2 = x + z
+
+new_op = OpFromGraph([x, y, z], [out1, out2])
+# This apply new_op to x, y and z. outs is a list of output variable.
+outs = new_op(x, y, z)
+f1 = theano.function([x, y, z], [out1, out2])
+f2 = theano.function([x, y, z], outs)
+
+print "Normal graph"
+theano.printing.debugprint(f1)
+print "Encapsulated graph"
+theano.printing.debugprint(f2)