Fix performance issue for large WACZ/WARC files on a remote location

[leewesleyv]

Resolves #21

• Make sure the tmpfile is removed at the end

Extract the WARC data instead of streaming data which was a bottleneck for remote WACZ files. Ran this on a 2.5 GB WACZ file with a spider that makes 20266 requests and the elapsed time came out at ~591 seconds. Whereas this will currently take around 20266 * 110 seconds. However, this does introduce some delay at the start since it extracts the WARC information into a local tmp file.

[wvengen]

Super, great idea to just cache the index 👍
I've experimented in the past with downloading the whole file first, but only the index seems much more sensible!