Improve RelateNG performance for A/L cases in prepared predicates

[dr-jts]

Improve the RelateNG self-noding check to allow A-L cases to take advantage of prepared predicates.

Previously, A/L cases were not using the prepared noder cache for predicates which require self-noding (such as contains, covers, relate etc). This caused significantly poor performance for a common situation. The self-noding check has been refined to allow A/L cases to be use the cached noder.

This makes RelateNG in prepared mode as fast as PreparedGeometry for all cases (and it handles more cases).

For example, a test in RelateNGPolygonLinesOverlappingPerfTest has dramatically reduced execution time (133x faster):

-------  Running with A: polygon # pts = 200000   B # pts = 400  x 10000 lines
runContainsOld : 9888 ms
runContainsOldPrep : 36 ms
runContainsNG : 5906 ms

BEFORE: runContainsNGPrep : 4540 ms
NOW:    runContainsNGPrep : 34 ms