Application callback.

Start the application before you load or instantiate anything: it owns the module cache and the node page budget.

The tree is one supervisor per subsystem, so a restart budget is shared only by things that belong together; wasm_sup draws it. It was flat until a stress run showed what that costs: five servers under one intensity => 5, period => 10 meant losing the module cache repeatedly could take the engine and the keeper with it, and the tables went too. wasm_engine still starts before anything that allocates, because node-wide page accounting must be in place before linear memory can grow: atomics pages are off-heap and invisible to max_heap_size, so nothing else would bound them.