Modify collators to not produce blocks if there's an upcoming session change at the next relay chain block

[alindima]

Suggested in #5461 (comment)

It makes sense, because cores are freed on session change anyway so these candidates have no chance of being included.

[burdges]

Interesting, well fair enough.

[bkchr]

I'm still in favor of just dropping this requirement of dropping candidates on session boundaries. If the validator set doesn't change, there is no need to drop candidates.

Generally this issue is not really that useful. We have much more forks, which lead to "useless" blocks than we have session boundaries.

[burdges]

We've no other reshuffling at session boundaries that screws up backing group assignments or whatever?

We could obviously avoid dropping candidates on session boundaries, but maybe we've higher priorities right now? How much time around the session boundaries?

[bkchr]

Yeah for sure this is not important.