suggestion: Graceful shutdown instead of std::mem::forget + _exit(0)

[netrome]

Suggestion for #2489

Note: This is just a PoC for another shutdown mechanism to verify we can do graceful shutdown with the nearcore actors. It is primarily claude generated without much iterations from me so it shouldn't necessarily be merged as-is.

[kevindeforth]

Thank you for looking into this!

I did try to spawn and stop the actor system explicitly, but it still required a timeout to make the tests non-flaky, because actor_system.stop(); is non-blocking.
The big difference is that I didn't use

    near_store::db::RocksDB::block_until_all_instances_are_dropped();

which seems to solve this problem.

I have to look into this in more detail. I am also curious if the mpc node spawning it's own RocksDB instance might interfere with that block_until_all_instances_are_dropped function.

[netrome]

I have to look into this in more detail. I am also curious if the mpc node spawning it's own RocksDB instance might interfere with that block_until_all_instances_are_dropped function.

I don't think the mpc node RocksDB instance would interfere with the nearcore one. Afaik they are completely isolated from each other.