Fix IncompleteArchiveException when enumerating compressed tar archives via ArchiveFactory.OpenArchive

[Copilot]

ArchiveFactory.OpenArchive threw IncompleteArchiveException: Failed to read TAR header after the second entry when opening any compressed tar (.tar.gz, .tar.bz2, .tar.xz, .tar.zst). Uncompressed .tar files were unaffected.

Root cause

In StreamingMode.Streaming (used for all compressed tars), TarHeaderFactory.ReadHeader created a TarReadOnlySubStream wrapping each entry's data but never disposed it before looping back to read the next header. The underlying decompressed stream stayed positioned at the previous entry's data, not the next header block. The first directory entry (size=0) happened to work; any entry with non-zero data broke iteration of the next.

Changes

• TarHeaderFactory.ReadHeader / ReadHeaderAsync — Track the previous TarReadOnlySubStream in a local variable and Dispose/DisposeAsync it before reading the next header. TarReadOnlySubStream.Dispose skips unread data + 512-byte alignment padding; the existing _isDisposed guard makes double-dispose safe.

• TarArchive.LoadEntries / LoadEntriesAsync — Split LongName handling by mode. In streaming mode, the long filename is read immediately from header.PackedStream when the LongName header is received—before the foreach advances and ReadHeader disposes the packed stream. Seekable (uncompressed) mode retains the existing post-read behaviour.

Usage

// Previously threw IncompleteArchiveException after the second entry
using var stream = File.OpenRead("archive.tar.gz");
using var archive = ArchiveFactory.OpenArchive(stream);

foreach (var entry in archive.Entries)          // now enumerates all entries
{
    if (!entry.IsDirectory)
        entry.OpenEntryStream().CopyTo(destination); // read content during iteration
}

Note: compressed tars are streaming-only — entry content must be read in iteration order, not after collecting all entries up front.