Added RNNsearch.cnna in invalid tests, still subject to change

src/test/resources/invalid_tests/RNNsearch.cnna

+package RNNsearch;
+
+architecture RNNsearch(max_length=50, vocabulary_size=30001, embedding_size=620, hidden_size=1000){
+    def input Q^{max_length, vocabulary_size} source
+    def output Q^{max_length, vocabulary_size} target
+
+    layer GRU(units=hidden_size, bidirectional=true) encoder;
+
+    source ->
+    Embedding(input=vocabulary_size, output=embedding_size) ->
+    encoder ->
+    FullyConnected(units=hidden_size, flatten=false) -> encoder.output;
+
+    layer GRU(units=hidden_size) decoder;
+
+    1 -> OneHot(n=vocabulary_size) -> target[0];
+    encoder.state[1] -> decoder.state;
+
+    unroll<t> BeamSearchStart(width=5, max_length=50) ->
+    (
+        (
+            (
+                decoder.state ->
+                Repeat(n=max_length, dim=1)
+            |
+                encoder.output
+            ) ->
+            Concatenate(dim=2) ->
+            FullyConnected(units=hidden_size, flatten=false) ->
+            Tanh() ->
+            FullyConnected(units=max_length) ->
+            Softmax()
+        |
+            encoder.output
+        ) ->
+        Dot()
+    |
+        target[t-1] ->
+        Embedding(input=vocabulary_size, output=hidden_size)
+    ) ->
+    Concatenate() ->
+    decoder ->
+    FullyConnected(units=vocabulary_size) ->
+    ArgMax() ->
+    OneHot(n=vocabulary_size) ->
+    target[t] ->
+    BeamSearchEnd();
+}