Previously, only the total size of all DIMMs was used. For example, four
8 MB DIMMs would behave as a single 32 MB DIMM.
This change forces handling of the individual DIMM sizes.
We currently do not emulate shuffling of 4MB pages. For example, the
data of the first page of bank 1 does not move if you move it to a
different page.
The important parts of this algorithm are:
1) DIMM sizes can be determined.
2) The final memory allocation will be a single contiguous block.
Add supports_types to device descriptors so that we can determine
supported types without having to create the hardware device.
Add description so that each device can have an optional description.
