Support string datatype declaration for numpy array creation

[mha-py]

Hey!

When you have code like

arr = np.zeros((N, M), dtype='uint8')

you cant just put the jit decorator in front of the function because numba doesnt recognize the string declaration. What is working is from numba.types import bytes and use that type instead. Problem is, that this type declaration is only working with numba, not without. Means that numpy throws an error when there is no jit decorator anymore.
So i guess it would be cool if numba would recoginze string declarations of data type!

[gmarkall]

You can use the np.uint8 dtype (and other dtypes in the np module in general) to keep the code working with and without the decorator, e.g.:

from numba import njit
import numpy as np

@njit
def make_zeros_numba():
    return np.zeros(1, dtype=np.uint8)

def make_zeros_python():
    return np.zeros(1, dtype=np.uint8)

make_zeros_numba()
make_zeros_python()

compiles successfully.

[gmarkall]

Additionally, marking as a feature request for support for strings for the dtype name.

[esc]

@mha-py thank your for asking about this. I will close this as it is a duplicate of (the long standing issue) #2201.