Matrix like type for GIL

### Is your feature request related to a problem? Please describe.

Yes. Due to requests in #204 and #615 , it is clear that there is serious need for some sort of matrix like type that is not normalized by default. Image processing and related functionality is in a dire need of this.

### Describe the solution you'd like

#### C++ Example

```cpp
// just 2D matrix of floats
using matrix32f = boost::gil::image>;
using matrix32f_view = decltype(boost::gil::view(std::declval()));

template
float saturate_cast(GrayPixel pixel);
template
GrayPixel saturate_cast(float matrix_cell);

// often needed to convert to/from images read from disk
template
void value_preserving_copy(GrayImageView source, matrix32f_view destination);

template
void saturate_cast_copy(matrix32f_view destination, GrayImageView source);
```

### Describe alternatives you've considered

It is also possible to just implement the following:

```cpp
template
void normalize_copy(GrayImageView source, gray32f_view_t destination);

template
void denormalize_copy(gray32f_view_t source, GrayImageView destination);
```

### Additional context

I believe having just `matrix32f` will be enough for starters. Most of the CV and IP computations are done in floats anyway.