[drmemtrace] Incorrect treatment of split_value marker ordering in raw2trace

raw2trace has logic to "unread" entries and put them in a deque, which is required for some readahead logic during trace decomposition. The TRACE_MARKER_TYPE_SPLIT_VALUE markers hold the partial marker values that do not fit in one marker. This marker is expected to be present _before_ the actual marker. But raw2trace "unread" entry logic puts it back in the reverse order.