Implement a keras layer to round numeric values and truncate values with a max value.

[workingloong]

For numeric features, we may want to convert the values to integer id by rounding and then make embedding with those ids. Though we can implement the function by using bucketing with boundaries where 1 interval，it is tedious to set the boundaries and inefficient to transform if the max value is big.

Example：
inputs: [[1.2], [4.1], [6.7], [100.1]]
max_value: 50
default_value: 50
the outputs: [[1], [4], [7], [50]]