dtar: support creating archives in NFS

Unlike Lustre or GPFS, NFS is not as forgiving when writing a file with multiple processes. At at minimum when writing an archive file on NFS, we need to:

1. use an algorithm in which processes align file offsets on memory page boundaries
2. and we should use fcntl lock/unlock

After we compute the offset of entry headers (and data), we need to reassign work to processes based on page/block boundary alignment. In fact, different portions of an entry header may need to be written by different processes if it happens to span a page boundary. Also, the work to copy data chunks should be adjusted to align on page boundaries based on their offset within the archive file.

A similar problem exists for dbz2.

Start by tracing dtar's archive-writing path, especially offset computation, entry-header handling, and data-chunk copying. Study how the work is assigned across processes, then determine how NFS requires page or block alignment and fcntl locking; done means archive creation follows those rules, with the similar dbz2 problem addressed or explicitly scoped.