Deal with duplicated transitions in LTSs

[mlaveaux]

Malformed inputs could have duplicated transitions/edges in the LTS or PG inputs. The question is whether these should be considered errors, we fix them after reading (this can be expensive with a sort + dedup) or ensure that algorithms operating on them are robust against them. The latter choice might be valid now, but will that also hold for all future algorithms.

[mlaveaux]

In parity games we now remove duplicate transitions in reading them, assert that there are no duplicates in the constructor, and adapted the translation to produce no duplicates. This seems to be the most conservative, and correct, approach. However, for LTSs checking for duplicates, given their size, might already be complicated. Especially with the memory optimised builders removing duplicates is not really feasible.