Queries that currently match no archetypes still try to borrow archetypes that they _could_match

[cart]

so i found a kind of nasty bug related to the new parallel scheduler, but that led me to what should be a optimization for large projects

The issue is that the bevy_ecs Query knows more about its archetype access than the internal hecs Query. Lets say you want to query &mut T, but there are no entities with T. The bevy_ecs query knows that it doesn't need to access any archetypes this frame, so the scheduler schedules that system alongside other systems that might access &mut T archetypes. But then if you try to iterate the (empty) query in the first system, it still tries to borrow &mut T in all archetypes that have it
this is because hecs queries iterate over all archetypes and filter them as they go
awhile back i created this issue: #53 because if we have hundreds of archetypes then iteration becomes needlessly expensive. The "fix" here is to pass in a reference to the bevy_ecs Query archetype bitset, so it only accesses the archetypes it needs to this frame.

And if we're doing that to fix the borrow bug (panicing when two &mut systems both borrow T, even though only one actually needs it), we might as well use that data to optimize iteration too.

[cart]

resolved by: bd8e979