minor fix for the skiplist code, resulting in a false positive with valgrind, and in general into a useless small allocation

print warnings in redis log when a test raises an exception (very likely to be caused by something like a failed assertion)

DEBUG DIGEST implemented, in order to improve the ability to test persistence and replication consistency

very strong speedup in saving time performance when there are many integers in the dataset. Instead of decoding the object before to pass them to the rdbSaveObject layer we check asap if the object is integer encoded and can be written on disk as an integer.

added overflow check in the double -> long long conversion trick to avoid integer overflows. I think this was not needed in practical terms, but it is safer

If a float can be casted to a long long without rounding loss, we can use the integer conversion function to write the score on disk. This is a seriuous speedup

fixed a problem leading to crashes, as keys can't be currently specially encoded, so we can't encode integers at object loading time... For now this can be fixed passing a few flags, or later can be fixed allowing encoded keys as well

long long to string conversion speedup applied in other places as well. Still the code has bugs, fixing right now...

added specialized function to compare string objects for perfect match that is optimized for this task