CASSANDRA-21183 Safely regain ranges that were previously owned by a node

[alanwang67]

Currently nodes that are regaining ranges can result in subtle edge cases when we still have command stores left over from previous epochs that overlap with the range we are regaining.

To address this we:

1. Allow the regaining of ranges only if the epoch in which we previously owned the range is already retired
2. Mark as unsafe to read the ranges that we are regaining for the command store that previously owned the range and prevent future setSafeToRead invocations to revert this change.
3. Throw an error when we query two command stores for the same range

[belliottsmith]

I think this and block... aren't used, so could remove them (and the associated notify)?

[belliottsmith]

We probably don't need this; we can simply query redundantBefore, as this won't happen on any fast paths

[alanwang67]

Maybe I should come up with a better name, but this variable is for keeping track of the ranges that we marked as unsafe to read in the CommandStore because we are regaining the range for it, so it is a subset of the actual retiredRanges. I use this value to ensure that whenever we set ranges as safeToRead we exclude these ranges as it is no longer safe to read from them if there exists a CommandStore with a greater epoch that has the same range. If we query redundantBefore, I'm not sure if we will be able to get this value.