illumos gate

So, #6513 is nasty because even after the fix, on pool data written before the fix was applied, send|recv will still produce corrupted data on the recv side.

It was pointed out that a heavy-handed, but "correct", approach would be a toggle to ignore hole_birth metadata, but that leads to either ignoring the hole_birth data forever for correctness, or risking silent corruption on sends, neither of which sounds like a reasonable default.

Using a new feature (let's say "hole_birth_fix" for lack of a nicer name) in conjunction with the enabled_txg feature should permit ignoring only hole_birth data predating the "fix" feature being enabled - this would be read-compatible, and require marginal intervention on the part of people to enable once updated, but has the benefit of allowing safe use of the hole_birth feature while also using the conservative behavior to prevent silent corruption prior to the fix.