Internal: load_be/le should accept a BufferSlicer

[reneme]

Additional context here: #3716 (comment)

TL;DR: Reduce the boiler-plate when loading integers of a statically known bitlength from a buffer:

// currently
auto a = load_be<uint32_t>(slicer.take<4>());
auto b = load_be<SomeStrongIntegerType>(slicer.take<sizeof(SomeStrongIntegerType)>());

// then
auto a = load_be<uint32_t>(slicer);
auto b = load_be<SomeStrongIntegerType>(slicer);

Note that the BufferSlicer is currently part of stl_util.h which we may or may not want to include in loadstor.h. Perhaps it makes sense to move the BufferSlicer, BufferStuffer and concat() into their own dedicated header.