Reduce the size of the manual index by storing the entries in a more
compact form. This introduces the CompressedRef class, which is similar
to the DIERef class, but it is more compact because it lacks the
unit_offset field (which is not necessary now, as the unit is uniquely
identified by the rest of the fields). The user_id_t representation
would have the same size, but I did not use that here because:
- it's more complicated to convert it to/from a DIERef
- it's more clunky to handle as it is just an integer
The new class is for internal use only. It is converted to a regular
DIERef before returning it to the outside world.