[QA] Leaderboard for challenges is too slow

Problem Description

The overview page doesn't seem to work anymore. If data is requested, the requests seems to be running forever.

Analysis

The problem is caused by the leaderboard, which has been introduced for the QALD challenge. The way the leaderboard is generated (for each language and dataset, search for the best performance for each system) becomes too expensive for the data that we currently have in our database (22 languages, 35 datasets and 33797 systems).

Hotfix

• Disable the leaderboard

Solution

List ExperimentTask IDs that are relevant for the leaderboard of a challenge in an additional table

• When GERBIL is started with a challenge configured, it should run a CREATE TABLE IF NOT EXISTS query to create leaderboard table
• When an experiment finishes, a simple check whether the result is better than the best result of this system in the current leaderboard can be used to update the leaderboard if necessary
• When the leaderboard is requested, only the experiment tasks listed in the leaderboard table have to be retrieved.

No files or tests are named. Start by tracing GERBIL's challenge startup, experiment completion, and leaderboard request paths, then identify where the current per-language and per-dataset search runs. Done means the leaderboard uses a maintained table of relevant ExperimentTask IDs and the overview request no longer performs the expensive full search.