Additional checks required when recovering sectors in checkNextRecoveries

[Stebalien]

Currently, we call faultTracker.CheckProvable before declaring a sector recovered. However, this will pass as long as the sector is on-disk and is the correct size, even if it's actually corrupted.

Unfortunately, this means that, for corrupted sectors that still exist, we'll:

1. Declare them "recovering".
2. Try to prove them.
3. Fail, skip them, and retry the proof.

Unfortunately, at step 3, we only re-try 5 times at the moment (fixed in #4633). This is usually sufficient, but not when there are many corrupted sectors.

We don't usually checksum sectors before attempting to prove them because this would force us to read every sector from disk (proving a sector only requires reading part of it). However, in this case, we know that the sector is more likely faulty than not so we should perform a more expensive check.