Explore support for detecting duplicate issues

[chrmarti]

Exploration of existing ML approaches is needed.

Since we would eventually want to know the existing duplicate graph, that could also serve as a middle-ground from where we can continue with ML approaches or simply listing recent duplicate target issues.

[chrmarti]

Early testing with the Python package Gensim suggested that in 25% of cases of existing duplicates there would be the original (or another duplicate) among the first 5 candidates. Further testing showed that the hit rate drops quickly with the number of issues considered as candidates, more investigation in this will be needed for a more solid estimate. E.g., ignoring all issues closed for the current stable release might reduce the number of issues to consider.