The effect of natural selection is that the genes that propagate most are the ones that maximize the probability of that gene appearing future generations. There is no direct evolutionary pressure for an individual to behave in ways that benefit the species in general.
Yes, because there is no concept of "species" in nature, species is an arbitrary grouping we humans apply to the individuals and genetic lines that are actually involved in natural selection.
The female's genes are most likely to get passed down many generations if the males she mates with have genes that enable her male children to survive long enough to reproduce. Once a species gets into a state where females make a habit of eating the males during copulation, it can't readily evolve out of that state, because a female that is less aggressive will mate with males who can't succeed at mating with other females, leading to that line dying out. Sexual cannibalism is a local maximum.
So if at any point in the past there were evolutionary pressure that made cannibalism beneficial to the genes, we would expect to see it stick around until there's sufficient pressure against cannibalism to allow selection away from that local maximum.
> "a female that is less aggressive will mate with males who can't succeed at mating with other females"
That premise is the fundamental point that your argument hinges on and I see no reason to just assume it would be true, and can create plenty of plausible arguments (that I think are fairly self evident) for why it would be untrue.
But that is true regardless of whether a male is fit enough to survive a cannibalistic female, which means that pacifist females do not have an advantage in natural selection:
As long as there are enough males in circulation that can succeed in fertilizing before being eaten, then there's no pressure on females to drop the cannibalistic strategy. Any male that cannot win the cannibalism game can only fertilize pacifist females, but a male that can win the cannibalism game has genes that can be passed on to both cannibalistic and pacifist females. So the genes of cannibalistic females have broader reach within a species that has already gotten to majority-cannibalistic because her male offspring can fertilize both types of female rather than just one type.
Again, to GP's point, just because non-cannibalism would be more beneficial to the species as a whole doesn't mean that evolutionary pressure on individual genetic lines will select away from it. The individual line is often better off being selfish and selecting the strategy that gives them the broadest possible range of propagation (hence Richard Dawkins' title, The Selfish Gene), which in this case is cannibalism.
