add nansum api to math

[esythan]

PR types

New features

PR changes

APIs

Describe

add nansum api

x = np.array([[float('nan'), 0.3, 0.5, 0.9],  
                [0.1, 0.2, float('-nan'), 0.7]]).astype(np.float32)
x = paddle.to_tensor(x)
out1 = paddle.nansum(x)  # [2.7]
out2 = paddle.nansum(x, axis=0)  # [0.1, 0.5, 0.5, 1.6]
out3 = paddle.nansum(x, axis=-1)  # [1.7, 1.0]
out4 = paddle.nansum(x, axis=1, keepdim=True)  # [[1.7], [1.0]]

[paddle-bot-old]

Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

[jeff41404]

public api should be add in all list below

[jeff41404]

for paddle.Tensor.nansum, we should add nansum in list of tensor_method_func below

[Superjomn]

LGTM

[TCChenlong]

LGTM