libufdt: reduce the number of malloc/free calling

The performance of some bootloader malloc/free implementation isn't
optimized, but libufdt calls huge number of malloc/free to build a
real tree data structure.

The patch adds an ufdt_node_pool to reduce the number of malloc/free.
ufdt_node_pool asks a larger memory block in each time, and
distributes the memory into several ufdt_node (ufdt_node_fdt_prop
or ufdt_node_fdt_node). libufdt doesn't need to call malloc()
to allocate memory until out of larger memory block.

The setting of this patch is 1024 nodes in each memory block,
so it can reduce ~1023/1024 malloc/free calling times after the patch
is applied.

Bug: 32969430
Test: ./tests/run_tests.sh
Test: ./tests/run_performance_test.sh
Change-Id: Id95e74da1235b9e2fc306500686515ee6b93017d