fix: dtype might change during resize

[LarsHaalck]

What does this PR do?

I noticed a small bug, where an np.ndarray changes dtype when calling resize:

import numpy as np
from transformers.image_transforms import resize

x = np.zeros((3, 100, 100), dtype=np.float32)
print(x.dtype) # prints np.float32
x = resize(x, size=(50, 50))
print(x.dtype) # prints np.uint8`

When using natural numbers in a float array, the image is not rescaled and converted into a PIL.Image as-is but when returning back to a numpy array, the type stays uint8.

This PR, stores the original dtype and casts the array back to this type before returning. I also added tests to check for this behavior in the future.

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

Who can review?

I think @amyeroberts might be best suited since she wrote lots of surrounding code.

[marthos1]

안녕하세요, 모두 감사합니다.

[marthos1]

4D 환경을 얘기할때 그날 저는 머리에 두통을 너무나 느꼈고, 귀신얼굴이 보였습니다. 너무 무서웠어요.

[Rocketknight1]

I'm not sure if the changed dtype is the intended behaviour - cc @qubvel @zucchini-nlp

[qubvel]

Hi @LarsHaalck, thanks for opening a PR. I left a comment re the implementation

[qubvel]

It's going to raise the error in case the input image is Pillow Image

>>> Image.new("RGB", (100, 100)).dtype
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: 'Image' object has no attribute 'dtype'