GetAccountInfo can return incorrect reads?

[linuskendall]

I believe that getAccountInfo can return incorrect/invalid reads when accounts are written to on a fork that later doesn't get confirmed.

Suppose account XXX is written to on slot 123. This will create an entry in accounts table, with slot=123 and slot=123 will have status processed. However, if slot 123 is skipped / on a fork, then when slot 124 is rooted with parent 121 the state for account XXX in the accounts table will remain until it is written to next. This means that getAccountInfo(commitment=processed) will now return a state from a fork that has been discarded.

Proposed solution: The postgres-rpc-server should ignore any processed commitments that are on orphaned slots.