Add more datatypes in convert DataFrame to numpy

[hkvision]

Currently many datatypes are not correctly converted, and may cause error in subsequent model forward.

[hkvision]

failed http://10.112.231.51:18888/job/BigDL-Orca-PR-Validation/590/

[shanyu-sys]

LGTM

[hkvision]

In our unit tests, many of the code looks like this:

df = rdd.map(lambda x: (np.random.randn(50).astype(np.float).tolist(),
                                [int(np.random.randint(0, 2, size=()))])
                     ).toDF(["feature", "label"])

Since python only have float (default is float64), int (default is int64), when converting to DataFrame, the schema will be array and array. Then in this case the model will throw error RuntimeError: expected scalar type Float but found Double

A few unit tests for example test.bigdl.orca.learn.ray.pytorch.test_estimator_pytorch_backend.TestPyTorchEstimator testMethod=test_data_parallel_sgd_correctness throws the opposite error RuntimeError: expected scalar type Double but found Float, which is really weird, probably a bug of error report in PyTorch...

@yushan111 Any comments on this?

[hkvision]

Comment out the data conversion for array[float/double] and jenkins passed: http://10.112.231.51:18888/job/BigDL-Orca-PR-Validation/599/