For some reason compilers other than GCC PowerPC 4.5.8 were not complaining
about the `use_shadow` being (re)defined outside the `Shadow` structure
definition.
This closes#2.
The inlining issue was due to a bad template design in the access policies
implementations: the MMIO_t type from the field was used (which is a volatile
type) for all arguments in the write implementation.
The new implementation relies on two template types (one for the register memory
device and another one for the value to be written). This fixes inlining and was
checked on godbolt.
