Add support for two new datasets: Thunderbird and OpenStack.

[dhonza]

Add Support for Thunderbird and OpenStack from here: https://github.com/logpai/loghub

• Thunderbird

• OpenStack

[dhonza]

The problem with all but the HDFS dataset is that these are labeled in a per-log line fashion, while the annotation is typically based on simplistic logline features such as detecting the phrases "ERROR" or "WARNING". We found the problem with the BGL data which makes that dataset practically unusable for our purposes. Any method working with the actual semantics of loglines, such as our methods based on global or contextual embeddings can detect similar phrases easily.

Algorithms based on template extraction (e.g., as https://github.com/logpai/loglizer) have a much harder time on these datasets. Nevertheless, we are more interested in detecting "harder" anomalies which are represented by more complex interplay of log messages.