This PR makes some improvements to the log downloading functionality of RLA, which I had to use to prepare the GHA training data. The two changes are:
Urlencode the file names to avoid problems when job names include special chars.
Run the log compression in the threadpool to avoid the download taking ages (compressing a single log took 30 seconds on my machine, and to train we need hundreds of them).
This PR makes some improvements to the log downloading functionality of RLA, which I had to use to prepare the GHA training data. The two changes are: