This fixes a memory issue. When you assign to base_iodelay_rules from itself, the LHS when you do `base_iodelay_rules[id_DELAYA] = base_iodelay_rules[id_DELAYB]`, can cause a heap allocation which possibly invalidates the memory of the RHS.
This was found while running nextpnr under ASAN. It might be useful to add testing under libasan into CI.