Avoid unnecessary load of index segment chunks during first creation

org.infinispan.lucene.InfinispanIndexOutput.getChunkById(AdvancedCache, FileCacheKey, int, int, FileMetadata)
is performing a get() operation for buffers even when it can know from the filesize that such an element will not be found. This is significantly slow when a CacheStore is used, as the cache won't have the value for sure and so it will attempt to load from the store.