Simplify Fluid Book Unit Tests

[abhinavarora]

A very common issue that users of Fluid find are that our book unit tests (https://github.com/PaddlePaddle/Paddle/tree/develop/python/paddle/fluid/tests/book) are not very simple for new users. Our book unit tests have tests for CPU/GPU, train/inference and single machine/distributed in the same file. This makes the file very lengthy.

Also, some tests have multiple models in the same file such as test_recognize_digits.py. This has MLP and Conv models in the same file and can end up confusing users. Hence we should make sure that we split all the tests into more granular self-contained tests. This will make it easier for new users to understand Fluid.

[abhinavarora]

• test_fit_a_line
• test_image_classification
• test_label_semantic_roles
• test_machine_translation
• test_recognize_digits
• test_recommender_system
• test_understand_sentiment
• test_word2vec