Allow open indices to be restored

[Tim-Brooks]

Currently, open indices cannot be restored. However, in CCR we have a
use case where we would like to restore an follower index if it has
fallen too far behind a leader index. This commit allows a
restoreOpenIndex parameter to be passed to the restore request. If
this parameter is true, an open index can be restored. This introduces
some concurrency issues (multiple attempts to restore the same index).
So this PR fails a restore request if there is already an in-process
restore for the requested index.

[elasticmachine]

Pinging @elastic/es-distributed

[Tim-Brooks]

@ywelsch - I know that we had discussed making restore requests idempotent. But that looks a little tricky to me? We would need to validate that the snapshot, includeAliases, ignoredShards, and other request parameters are the same? Otherwise the requests would have produced different restores. It seemed safer to just reject another request if there is one in progress. Although I can adjust and attempt to implement all the validation (that the different request parameters are the same) if you think it is necessary functionality.