[SeiDB] Fix MemIAVL Race Condition during snapshot reload

[yzang2019]

Describe your changes and provide context

Problem:
In EVM, serving EVM RPC endpoints will have to query and access the latest state or historical state, when accessing latest state from MemIAVL, there's a race condition where the key we are trying to access doesn't exist because in the background, we have already switched to a difference snapshot after snapshot rewrite completes. Here's how it could happen:

1. Goroutine A: get a Tree pointer, preparing to call tree.Get
2. Goroutine B: reload snapshot and switched to a new tree, closed old tree
3. Goroutine A: Still use the old tree pointer to call get, got segmentation fault

And here's the error when this race condition is triggered:

Solution:
This PR address the issue by doing an internal replacement, instead of closing the old tree and create a new tree, we replace all the variables within each tree in a thread-safe manner as well as clean up and closing the old snapshots safely, this guarantees that all other operations will be block while we are replacing the whole tree.

Testing performed to validate your change

Add Unit Test to repro and catch race conditions

[yihuang]

Iterator is tricker, it supposed the hold the RLock before the iterator is closed

[yzang2019]

Hmm yeah that's a good point, iterator is a bit tricky, this lock only covers the creation of the iterator, might not be enough. I think for now it's probably fine since we don't have a lot of use cases using iterators in EVM, we definitely can add more locking coverage in the future

[alexanderbez]

Changes LGTM.