In general it looks good to me, except we never explicitly talked about adding MonadFail as a default export to Control.Monad or Prelude in this release.
Given the lengthened timeline it might be better to warn about the fact that they are coming into base next release than to take symbols now?
My proposal would be to remove the re-exports of MonadFail (sans method) from Prelude and Control.Monad, add a warning that they are going to be added, and then bring them in in 8.2 or so.
Looks good to me.
When I wrote this as part of the MFP implementation, I was wondering about the INLINE pragmas in the default MonadFail implementations. They seem pretty redundant, since given the size of the functions GHC should easily be able to decide whether to inline the definitions or not.