Making TupleMerge a Runtime Option and Validating Correctness

[mjgiardino]

Right now, whether the system uses block merging (i.e. an entire block is removed from the AntiCacheDB and all its tuples merged into the in-memory table) or tuple merging (i.e. a single needed tuple is merged into the table and the block remains in the AntiCacheDB), is determined by setting variables in persistenttable. This should be a runtime option, using an option in HStoreConf.

In addition, tuple merge needs to be validated for correctness both via code review and by adding test cases. Bugs have been fixed in the fork mikelj/h-store (commits: mjgiardino@cc86df2, mjgiardino@6cd058c, mjgiardino@5848065) but there may be more errors.