Investigate the benefit of more precise soft ref policy after marking

Since JDK-8269596, the soft ref policy is set up before marking, and the memory pressure is derived based on the heap usage at the end of previous GC cycle. A more precise measurement of memory pressure is to use the live set size after marking. This also means an additional "soft ref reconsideration" subphase in ref-processing phase. This new subphase could involve marking a large objects graph, so it's probably too expensive for STW collectors.