Fixed a singular matrix occurrence problem.

When the same inverse location problem was performed several times in
raw, the same result was added several times in the cache. This lead to
singular matrices being created as sample points were not independent.

src/main/java/org/orekit/rugged/linesensor/SensorMeanPlaneCrossing.java

@@ -419,12 +419,22 @@ public class SensorMeanPlaneCrossing {
             }
             if (FastMath.abs(deltaL) <= accuracy) {
                 // return immediately, without doing any additional evaluation!
-                if (cachedResults.size() >= CACHED_RESULTS) {
-                    cachedResults.remove(cachedResults.size() - 1);
+                final CrossingResult crossingResult =
+                    new CrossingResult(sensor.getDate(crossingLine), crossingLine, target,
+                                       targetDirection[0], targetDirection[1]);
+                boolean isNew = true;
+                for (final CrossingResult existing : cachedResults) {
+                    isNew = isNew && FastMath.abs(crossingLine - existing.crossingLine) > accuracy;
                 }
-                cachedResults.add(0, new CrossingResult(sensor.getDate(crossingLine), crossingLine, target,
-                                                        targetDirection[0], targetDirection[1]));
-                return cachedResults.get(0);
+                if (isNew) {
+                    // this result is different from the existing ones,
+                    // it brings new sampling data to the cache
+                    if (cachedResults.size() >= CACHED_RESULTS) {
+                        cachedResults.remove(cachedResults.size() - 1);
+                    }
+                    cachedResults.add(0, crossingResult);
+                }
+                return crossingResult;
             }
             for (int j = 0; j < i; ++j) {
                 if (FastMath.abs(crossingLine - crossingLineHistory[j]) <= 1.0) {