Global Graph Updates Failing in Extended Runs

For a while there has been an issue with Dash sometimes skipping global graph updates. Previously this has generally been fairly minor: if it misses an update, it will update to the latest graph within two or three blocks, causing only a short-term delay.

While testing the most recent branch, a much more serious version of this behaviour has appeared. After a certain point in an extended simulation, global graph updates will cease entirely. Pausing will not fix it. Resetting the simulation also will not fix it, despite the "reset_simulation" callback returning empty graph updates to all graph views. The beginning of the next simulation will display the loading animation, but as soon as graph "updates" start appearing, the stuck update from the previous run will be the only one displayed. It persists through multiple cycles of resetting and restarting. Waiting for several minutes while paused or after restarting had no effect. Increasing the minimum delay between updates seemed to delay the occurrence until later block numbers, but not solve it.

Curiously, miner graphs do no seem to be affected at all. They have been synched with the state of the trial even after the global graph has been stuck for many blocks and through multiple restarts.

Reproduce the failure during an extended simulation, then trace the reset_simulation callback and the global graph update path. Compare global graph updates with the still-synchronized miner graphs; done means global graphs continue updating correctly across pauses, resets, and restarted simulation cycles.