Isolate corrupted BLOBs when corruption is detected at runtime

[ikopylov]

Currently, corrupted BLOBs isolation logic is only executed at startup and its checks are limited. It can only detect real issues if the index regeneration is required and thus a full BLOB scan is performed. Performing a full scan of all BLOBs at startup will be extremely slow, so the better solution would be to detect corruption at runtime, when read operation fails.

[ikopylov]

Also detect corrupted blob during index regeneration (If it hasn't been implemented yet)