Force features to be floats instead of doubles

As per @caitlynlee in google-research#7 , Make predictions give errors for various tensors as `expected dtype double does not equal original dtype float`. Forcing tensors  to be floats instead of doubles in `astronet/predict.py ` solves the issue.

```
115: global_view =  preprocess.global_view(time, flux, FLAGS.period).astype(np.float32)
...
120: local_view = preprocess.local_view(time, flux, FLAGS.period, FLAGS.duration).astype(np.float32) 
```

exoplanet-ml/astronet/predict.py

@@ -112,12 +112,12 @@ def _process_tce(feature_config):
   features = {}
 
   if "global_view" in feature_config:
-    global_view = preprocess.global_view(time, flux, FLAGS.period)
+    global_view = preprocess.global_view(time, flux, FLAGS.period).astype(np.float32)
     # Add a batch dimension.
     features["global_view"] = np.expand_dims(global_view, 0)
 
   if "local_view" in feature_config:
-    local_view = preprocess.local_view(time, flux, FLAGS.period, FLAGS.duration)
+    local_view = preprocess.local_view(time, flux, FLAGS.period, FLAGS.duration).astype(np.float32)
     # Add a batch dimension.
     features["local_view"] = np.expand_dims(local_view, 0)