Result level cache key collisions from utf8 encoding

[gianm]

The ResultLevelCachingQueryRunner does this when it computes the cache key for a query:

final String cacheKeyStr = StringUtils.fromUtf8(strategy.computeResultLevelCacheKey(query));

This is bad because not all byte arrays can be represented as Java Strings. For example, the byte arrays [63, -48, 0, 0, 0, 0, 0, 0] and [63, -24, 0, 0, 0, 0, 0, 0] have the same representation when converted to Strings. Both have an invalid second character, which both get mapped to the replacement character. These byte arrays are the representations of doubles 0.25 and 0.75 respectively, which is how this was originally noticed (two queries had a cache key collision when they differed only in that one used 0.25 and one used 0.75 as a quantile parameter).

To fix this we need to always use byte[], never String, when dealing with cache keys.