Use the faster bit interleave for encode_quadkey itself

encode_vertex() computed exactly the same thing as encode_quadkey(),
so there is no reason to have both. Give encode_quadkey() the branch-free
implementation, which also speeds up the default encode_index, and have
the shared node code call it directly. The unit test now compares it
against the old bit-at-a-time loop and checks that decode_quadkey()
reverses it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P2nBqZisxNQfmEmon3vE9v

main.cpp

@@ -2122,7 +2122,7 @@ std::pair<int, metadata> read_input(std::vector<source> &sources, char *fname, i
 #endif
 
 				struct node n;
-				n.index = encode_vertex((unsigned) x, (unsigned) y);
+				n.index = encode_quadkey((unsigned) x, (unsigned) y);
 
 				fwrite_check((char *) &n, sizeof(struct node), 1, readers[0].nodefile, &readers[0].nodepos, "vertices");
 			}