Use single hash tables for compoment insertion/removal edges to improve spatial locality.

[adamreichold]

No description provided.

[Ralith]

Very tidy, thanks! Out of curiosity, any changes to the benchmarks? We're probably not exercising cases where this would matter (very high cache pressure, perhaps), but maybe it registers anyway.

[adamreichold]

Out of curiosity, any changes to the benchmarks? We're probably not exercising cases where this would matter (very high cache pressure, perhaps), but maybe it registers anyway.

I added a benchmark stressing insertion and removal but sticking around a steady state of attached components. This yields

test insert_remove                  ... bench:         356 ns/iter (+/- 225)

before and

test insert_remove                  ... bench:         330 ns/iter (+/- 123)

after these changes. I'd say close to noise with a positive tendency?

[Ralith]

Thanks for investigating!