[examples] Use original eps value for MNIST robustness.

doc/mnist.rst

@@ -75,7 +75,7 @@ The CAISAR interpretation language `caisar.mlw
 We will import the relevant theories with the ``use`` keyword.
 As described in :ref:`interpretation`, the ``Vector`` theory provides
 a vector type, a getter (``[]``) operation and a ``valid_index`` predicate
-that determine whether the get operation is within the range of the vector length.
+that determines whether the get operation is within the range of the vector length.
 ``NeuralNetwork`` defines a type and an application function (``@@``).
 We will also need integers and floating point numbers
 to declare and define :math:`\epsilon`. 

examples/mnist/property.why

@@ -34,6 +34,6 @@ theory MNIST
   goal robustness:
     let nn = read_neural_network "nets/MNIST_256_2.onnx" ONNX in
     let dataset = read_dataset "csv/mnist_test.csv" CSV in
-    let eps = (0.375:t) in
+    let eps = (0.0100000000000000002081668171172168513294309377670288085937500000:t) in
     robust nn dataset eps
 end