Targetran

- Python 3.7 only
- Fix instantiating TF random number generator: do not use from non deterministic state

Breaking changes:
- Use newest `numpy.typing` feature to strengthen the type annotations of the NumPy related code, see [here](https://github.com/bhky/targetran/blob/ebf19b71a97361f1adfcbe71dd2586ea9371d95b/README.md#notations) for the notations.

Update information dunders.

Python requirement updated. It now requires python>=3.7.

Improve code speed for `BILINEAR` interpolation.

Add `interpolation` mode option to:
- `CombineAffine`/`TFCombineAffine`
- `RandomRotate`/`TFRandomRotate`/`rotate`/`tf_rotate`
- `RandomShear`/`TFRandomShear`/`shear`/`tf_shear`
- `RandomTranslate`/`TFRandomTranslate`/`translate`/`tf_translate`

Currently, `BILINEAR` (default) and `NEAREST` are supported.