libufdt: device tree overlay via unflattening FDT

The original version of libdtoverlay is slow in searching for
particular nodes and adding subnodes/properties due to the operations
on flattened device tree (FDT).

`libufdt` builds a real tree structure (named ufdt -- unflattned
device tree) from FDT. In the real tree, we can perform certain
operations (e.g., merge 2 subtrees, search for a node by path) in
almost optimal time complexity with acceptable additional memory usage.

With libufdt, we improve the merging of two dtb files from O(N^2) to
O(N), where N is the number of nodes in the tree.

Bug: 30800619
Test: run test scripts (see libufdt/tests/README)
Test: manually ported libufdt into a bootloader,
      checked it can merge a base dtb and a dtbo to generate a final dtb
      to boot the device

Change-Id: I1ff886bb8a62bad1451edcd7c4fe5cb48b7a034a