Clean up UBSan integer false positives

This is #108 follow-up.

### Problem

For example, `variant=ubsan_integer` failures like in [build 375.9](https://travis-ci.org/boostorg/gil/jobs/396282494#L1181), may be revealing some issues regarding actual unsigned type used in conversion routines.

The following UBSan integer runtime error

```
channel_algorithm.hpp:344:64:
runtime error: unsigned integer overflow:
4294967287 + 128 cannot be represented in type 'unsigned int'
```

It is reported for `channel_convert_to_unsigned` utility

https://github.com/boostorg/gil/blob/d882b844944001f5f94e55aed3faaf651aae3032/include/boost/gil/channel_algorithm.hpp#L339-L346

For `val = -9` input, the result is calculated as:

```
((4294967295 - 8) + 128) % 256 = 119
```

Obviously, changing the intermediate conversion to
`static_cast(val)`
or
`static_cast(val)`
yield equivalent results:

```
((65535 - 8) +128) % 256 = 119
((18446744073709551615 - 8) + 128) % 256 = 119
```

However, it may be a good idea to inspect all the `ubsan_integer` failures for any unexpected results.