[Sparse]add sparse unary api(sin/tan/pow/neg/log1p/square/cast...)

[zhwesky2010]

PR types

New features

PR changes

APIs

Describe

增加SparseTensor一元操作API

1. paddle.incubate.sparse.sin(x, name=None)

import paddle
from paddle.fluid.framework import _test_eager_guard

with _test_eager_guard():
    dense_x = paddle.to_tensor([-2., 0., 1.])
    sparse_x = dense_x.to_sparse_coo(1)
    out = paddle.incubate.sparse.sin(sparse_x)

2. paddle.incubate.sparse.nn.Relu() (x, name=None)
3. paddle.incubate.sparse.tan(x, name=None)
4. paddle.incubate.sparse.asin(x, name=None)
5. paddle.incubate.sparse.atan(x, name=None)
6. paddle.incubate.sparse.sinh(x, name=None)
7. paddle.incubate.sparse.tanh(x, name=None)
8. paddle.incubate.sparse.asinh(x, name=None)
9. paddle.incubate.sparse.atanh(x, name=None)
10. paddle.incubate.sparse.sqrt(x, name=None)
11. paddle.incubate.sparse.square(x, name=None)
12. paddle.incubate.sparse.log1p(x, name=None)
13. paddle.incubate.sparse.abs(x, name=None)
14. paddle.incubate.sparse.pow(x, factor, name=None)
15. paddle.incubate.sparse.cast(x, index_dtype=None, value_dtype=None, name=None)
16. paddle.incubate.sparse.neg(x, name=None)

增加乘法、除法scalar的操作

1. paddle.incubate.sparse.multiply(Tensor, scalar)
2. paddle.incubate.sparse.divide(Tensor, scalar)

增加类成员函数nnz

1. paddle.Tensor.nnz()：返回SparseTensor中非0元素的个数

[zkh2016]

这里可以支持下float16

[zhwesky2010]

thx~下个PR改一下

[XiaoguangHu01]

LGTM