test_nn_dropout cases fail to validate drop_prob correctly

[wong-1994]

When implementing nn.Dropout (training mode), either passing self.p or (1 - self.p) to init.randb function can pass test_nn_dropout's cases.

Although test_nn_dropout_forward_1 and test_nn_dropout_backward_1 pass 0.45 and 0.26 as the prob parameter respectively, under the current random seed, the array generated by the init.rand function always has 50% of values less than prob and the other 50% greater than prob, leading to the above result.

Specificlly, output of init.randb function would be:

test_nn_dropout_forward_1 :
[[ True False False]
[ True False True]]

test_nn_dropout_backward_1:
[[ True False False]
[ True False True]]

due to output of init.rand function being:

test_nn_dropout_forward_1 :
[[0.01457496 0.91874701 0.90071485]
[0.03342143 0.95694934 0.13720932]]

test_nn_dropout_backward_1:
[[0.01457496 0.91874701 0.90071485]
[0.03342143 0.95694934 0.13720932]]