fix(editor): the track shown while it's laid over the ground no longer holds up the frames: its grade line is fitted on a thread of its own, the line fitted a moment ago stretched over the track meanwhile, and the click takes the line fitted for the same ground instead of fitting it again; the terrain follows the tracks once the laying ends, not after each click of a chain

Fitting the grade line took 3 ms for 250 m of track, 30 ms for 1 km, 65 ms for 2 km and 200 ms for 4 km, and the preview of the lay tool and of the drag from a free end fitted it every frame, twice for the drag; on the terrain model the preview of some 800 m took 10 to 17 ms a frame, the cursor standing or not. Now the frame waits under 1.5 ms for 4 km. The formations of the tracks led into the terrain went through every chunk under them 0.3 s after each click, some 50 ms, while the cursor went on to the next point; they wait for the laying to end. The tracks laid and the terrain are the same as before.

Selftests: perf logs the longest frame and those over 33 ms, fittime the time of the fit by length, terrainhash a hash of the chunks in memory.

application/editormodeselftest.h

@@ -24,6 +24,19 @@ inline std::size_t &frames()
 	return count;
 }
 
+// the longest frame since the last perf, and the frames longer than 33 ms
+inline double &frame_longest()
+{
+	static double value{0.0};
+	return value;
+}
+
+inline std::size_t &frames_slow()
+{
+	static std::size_t count{0};
+	return count;
+}
+
 inline std::string one_line(std::string Text)
 {
 	std::replace(Text.begin(), Text.end(), '\n', ';');