Sanely handle repeated runs

[dstufft]

So right now pypi-cdn-log-archiver basically assumes it's only going to be ran once per day. However this might not hold true, especially with the changes to skip failed logs that will hopefully be done in the future. In order to protect against data loss it should either refuse to run if the target .xz file already exists (and thus it would overwrite) or it should pull down that .xz and concat any additional logs to it.

[benjaminws]

+1, I'll work on this!

[benjaminws]

@dstufft How would you feel about refusing to upload if the archive already exists on the remote end, and is the same size as the archive to be written. In other cases it would just overwrite the destination like it does now.

I don't mind the concatenation route, but if I'm honest, it's easier to just move on if the archives are not the same name/size.

[dstufft]

Probably it should just refuse to run?

[benjaminws]

Probably :)