Refactor github archive downloader

[radiohead]

What

Refactor github archive downloader with the following notable changes:

1. Stop reading full archives in memory. This was unnecessary, since the code is using buffered IO reader anyway. The downside here was huge memory consumption, since individual archives were taking around 200 MB in RAM.
2. Add optional flag to allow marking files as downloaded even if some events had errors.
3. Wrap DB operations in transactions - otherwise it was possible that DELETEs would succeed but INSERTs would fail.
4. Use context.Context instead of "done" channels for timeouts and cancellations.
5. Migrate to Go 1.13 so that we can use http.NewRequestWithContext.

Why

The downloader job is using so much RAM it's getting OOMKilled. I couldn't even download a week worth of files with 4 workers, because the container was using ~4 GBs of RAM (which was the limit on my Docker VM on macOS).

Some files contain events that cannot be parsed due to broken (?) JSON contents (e.g. invalid character '\\'' looking for beginning of value). Those files will never be marked as downloaded, so skipping broken events is the only way to mark those files as downloaded - hence the optional flag. We could still have the flag set to false (as it is by default) and the functionality won't be changed.

Transactions are nice to have, although I haven't noticed inconsistencies in the DB during my test runs. The rest of refactoring was done to make the code a bit more readable - feedback is welcome.

[marefr]

LGTM great work. Haven't tested it though, but looks good