vdpa_sim_net: check TX pull result before RX copy

In the Linux kernel, the following vulnerability has been resolved: vdpa_sim_net: check TX pull result before RX copy vringh_iov_pull_iotlb() returns a signed byte count. A failed TX pull is currently added to the unsigned byte counter and then passed as a size_t length to receive_filter() and vringh_iov_push_iotlb(). A negative error can therefore become a large length in the RX path. Handle non-positive pull results before every length use. Count the TX error and complete the consumed TX descriptor with zero bytes. I found this bug myself, though the patch was written with AI assistance.