Use LZMA compression for downloading from FileSrv

[dpogue]

I believe right now it uses .gz files with plZlibStream. 7zip offers better compression, especially if we can combine multiple files into a single archive.

Future ideas based on this support:

• Compression of Python and SDL downloaded from server. Download a single archive that unpacks to the required multiple files.
• Manifest-based archives. Rather than downloading each file, you could download all the files for an Age as a single archive.

[Deledrius]

Would it be best to use LZMA SDK directly (http://www.7-zip.org/sdk.html) or something like libarchive (http://code.google.com/p/libarchive/) which looks like it has a better interface, as well as giving us a lot of extra formats for free? I'm thinking it might be nice to implement a plCompressedStream and let the lib handle formats entirely (as a bonus, it would allow us to fold plZlibStream into it as well, further cleaning up the glut of sub-projects).

[zrax]

zlib != gzip, and LZMA != 7z... gzip and 7z are file formats, whereas zlib and LZMA are stream formats... Just picking the nits :P

Anyway, I believe we'll still want the latter here, since it means individual files can be downloaded when only one is out of date or corrupt, instead of having to download an entire archive. The exception is, of course, the Python/SDL, since the current engine design downloads those every time the game is launched...

[zrax]

An ancient issue, but something else to consider here is algorithm performance. LZMA offers better compression, but requires significantly more processing power to compress and decompress. On the other hand, something like zstd offers slightly better compression than zlib, but with notably better decompress speed, and lz4 offers slightly worse compression but with massively improved speed.

[Hoikas]

With some of the dinosaurs folks play URU on, decompress speed should be considered the most important metric IMO. Considering the results of zlib compression on the assets (it's pretty good), I'm not even sure investing time into a new algorithm (and properly supporting backwards compatibility with old client/servers) is particularly desirable.

[dgelessus]

(see also #737)