Optimized RocksDBStore.IterateIndexes() for pruned chain storage.

[greymistcube]

See #4003 for context.

Should follow #4001.
This assumes a transition from a pruned storage to an unpruned storage is never allowed.
Together with #3999 and automatic pruning in RocksDBStore() constructor and further removal of chain id manipulating methods, the implementation should become much simpler.

I haven't tested but I don't think Iterator.SeekToPrev() would be any faster. This is still slower than the original bugged implementation of IterateIndexes(), but the performance degradation is much more tolerable.

[greymistcube]

Added 49ae37f for further optimization.

The point is to terminate iteration early at the exact point it should terminate.
Although not a proper benchmark, a preliminary benchmark shows a vastly improved indexing speed.

For unpruned Odin chain:
Took 28 ms to access last 1000 hashes

For pruned Odin chain:
Took 2 ms to access last 1000 hashes

Just another reason why pruning should be done. 😗

[greymistcube]

Unless we want to support partial chains, the removed test is invalid.