I guess you could check past contests to see what are the most popular algorithms/concepts to use (namely Breadth first search, Depth first search, Beam search, Genetic algorithm, Hill climbing, Simulated annealing, Voronoi, Monte carlo, Monte carlo tree search, minmax, minmax with itérative deepening). I don’t think you need to go deep for each but merely understanding their concepts should enable you to know which one has more potential for the contest (and if you don’t manage to pick one, you can always ask in the chat if it’s a good idea to pick this or that algorithm). Then during the contest you are likely to have enough time to implement the one you choose.
I would also recommend to at least train on a “discrete” multi (one with small grids and cells) and one with physics simulation (coordinates are still discrete but they are rounding, speed/acceleration/collision/bouncing considerations as well -> coders strike back, fantastic bits, poker chip race, mena max…). According to me, while discrete multis are rather “natural”, multis implying physics are not that straightforward and could be annoying to start if you don’t have prior experience on it.