Refactor(nnx): Remove Redundant FFW Call in SharedExpertsTransformerBlock (#1089)

Author: sierraisland

tpu_inference/layers/jax/transformer_block.py

@@ -88,7 +88,6 @@ def __call__(self, x_TD, is_prefill, kv_cache, attention_metadata):
 
         if isinstance(self.custom_module, DenseFFW):
             dense_layer = self.custom_module
-            logits_TD = self.custom_module(normed_ffw_input_TD)
         else:
             dense_layer = self.dense_ffw