Replace the model used for the cache part of the peak-cache model with a simple vector of vectors. Avoids unnecessary locking in a class that is not thread-safe in any case. Also record whether the final column is actually truncated, rather than risk possible backward seeks to re-read it in the case where it simply might be