From Jeffrey Morlan: pst_build_id_ptr reads the Block BTree into a linked list, which pst_getID does a linear scan through. For large PSTs that have millions of blocks, this is extremely slow - almost all time is spent in pst_getID. Since the BTree entries must be in order, this can be dramatically improved by reading into an array and using binary search.

Need testing on big-endian machines. We may be missing some LEnn_CPU()
calls.

pst2diii needs header and mime type updates from readpst.

At the next soname bump (to libpst.so.5) we should
    move some of readpst into the shared library, in particular write_normal_email()