Hdiff fixes

[Inspector-Butters]

This PR addresses multiple issues in the hdiff code:

1. the function hasStateUsingStateDiff() that's called in db.HasState() was returning true for any slot that fell on the hdiff tree levels. This is wrong because those states might not actually be saved. This PR fixes that by actually looking in the db to check.
2. add the call to saveStateByDiff() in db.SaveState() to actually save the states in the hdiff tree. this was missed previously. the reason for the if guard s.stateDiffCache != nil over this call is that when syncing from genesis, the call to initialize state diff comes after the call of SaveState() to save the genesis state. which would cause a panic here. with this guard, that save goes to the legacy stateBucket. but the state will later get saved in the hdiff tree when initializeStateDiff() is called.
3. adds an early return to getStateUsingStateDiff() when computeLevel(slot) returns -1.
4. moves the call to initializeStateDiff() higher when doing checkpoint sync. specifically moves it before the call to SaveState(). otherwise there is a bug: state diff tries to save that state in the tree because it has offset 0 from the genesis call, then it would error out because the checkpoint slot doesn't have the necessary parents in the tree. (the offset isn't updated yet). this move makes it so the offset is updated before saving the state. the SaveState() call is also skipped after initializing state diff, because it's redundant. NOTE: maybe if this pattern is good, we want to do the same in genesis.go to avoid writing to the legacy stateBucket entirely. This is done in Hdiff fixes 2 #16534
5. other small fixes.

[prestonvanloon]

LGTM thanks