stop using r_eff in ps_tail_length

[avehtari]

I'm now convinced it was a mistake to use r_eff is ps_tail_length. Most of the time the difference is negligible, but sometimes it leads to strange behavior. I propose we drop it. We may need to keep allowing r_eff to be passed, but it should not be used for ps_tail_length computation. r_eff can still be useful for MCSE elsewhere

[n-kall]

Simple change, and r_eff argument can then be either removed and absorbed by dots or kept with a deprecation note in docs. Should we have some note in the documentation that it is updated and now deviates slightly from the PSIS paper?

[avehtari]

Yes, there should be a note in the documentation