Remove fast_run from mode in test_shape_i_const

`fast_run` includes the fusion optimization, but that should not be called with a `fast_compile` mode because that prevents the creation of `c_thunks`, even when a `C` compiler is available. This forces the use of Python perform method which is limited to 32 operands, when the Fusion Optimizer assumed that the C limit (1024 operands) was in place. In this test, one of the fused Composite surpasses that limit.

tests/tensor/test_subtensor.py

@@ -986,7 +986,7 @@ def test_adv_sub1_idx_broadcast(self):
     def test_shape_i_const(self):
         # Each axis is treated independently by shape_i/shape operators
 
-        mode_opt = self.mode.including("fast_run")
+        mode_opt = self.mode
         data = self.shared(np.array(np.arange(5), dtype=self.dtype))
         for start in [None] + [-8, -5, -1, 0, 1, 5, 8]:
             outs = []