Follow-up to #1012.

We're working on streaming large archives and we plan the work from the central directory first, then open readers only for the entries we actually need. by_index reads the local header for each entry and needs &mut, and by_index_data avoids that but still needs the ZipArchive<R> around. What we really want is to pass Arc<ZipArchiveMetadata> to other threads and read entries from it directly, without carrying a reader.

So this adds len, is_empty and entry on ZipArchiveMetadata. entry returns the same ZipFileEntry as by_index_data (and FileNotFound when out of range), and by_index_data / ZipArchive::len now just call into it. Added a couple of tests comparing entry(i) against by_index(i).

• The PR title conforms to Conventional Commits.