fix fp16 dtype checking for argmin op

[hua-zi]

PR types

Bug fixes

PR changes

APIs

Describe

问题描述：在静态图模式下，输入为FP16类型时，argmin会报TypeError。

import paddle
import numpy as np

paddle.enable_static()

x_np = np.random.random((10, 16)).astype('float16')
x = paddle.static.data(shape=[10, 16], name='x', dtype='float16')
out = paddle.argmin(x)

exe = paddle.static.Executor()
exe.run(paddle.static.default_startup_program())
out = exe.run(feed={'x': x_np},
            fetch_list=[out])

报错：

Traceback (most recent call last):
  File ".\test.py", line 11, in <module>
    out = paddle.argmin(x)
  File "c:\app\anaconda3\envs\pytorch\lib\site-packages\paddle\tensor\search.py", line 271, in argmin
    check_variable_and_dtype(
  File "c:\app\anaconda3\envs\pytorch\lib\site-packages\paddle\fluid\data_feeder.py", line 86, in check_variable_and_dtype  
    check_dtype(input.dtype, input_name, expected_dtype, op_name, extra_message)
  File "c:\app\anaconda3\envs\pytorch\lib\site-packages\paddle\fluid\data_feeder.py", line 147, in check_dtype
    raise TypeError(
TypeError: The data type of 'x' in paddle.argmin must be ['float32', 'float64', 'int16', 'int32', 'int64', 'uint8'], but received float16.

修复方案：在argmin API静态图模式下的类型检查中增加fp16支持