feat: add onStart/onStop/onSaved/onError callbacks to autosave options

[ascott18]

Why

Deep auto-saves attach and detach auto-save from entities as they enter and leave the object graph (new navigation properties, loaded collections, $addChild). Today there's no central place to observe the save lifecycle of each individual entity, so reacting to a successful/failed save or knowing when auto-save was attached to a nested entity requires wiring $save.onFulfilled/onRejected on every entity by hand.

What

Adds four optional callbacks to AutoSaveOptions (usable from both $startAutoSave and $useAutoSave, on item and list view models):

• onStart(vm) - invoked per entity as auto-save attaches, including entities attached to the graph after start.
• onStop(vm) - invoked when auto-save deactivates (explicit $stopAutoSave() or component unmount).
• onSaved(vm) - invoked after a successful auto-save $save.
• onError(vm, error) - invoked when an auto-save $save fails.

Each callback receives the affected view model, which is essential under deep since a single options object is shared across the whole graph.

Notes for reviewers

• onError fires only on actual $save rejections. Saves skipped due to client validation errors (already observable via $hasError) or a predicate do not invoke it, since no save is attempted.
• Consistent with existing behavior, $stopAutoSave() is not recursive: stopping a deep root only fires onStop for the root. On component unmount, onStop fires for every entity. This is documented in the callback doc comments and the ViewModels docs.
• Callbacks fire on actual entities, not on the ListViewModel container (a list never saves; its items each fire their own).

Includes 6 new unit tests covering start/stop/saved/error for both shallow and deep cases, plus CHANGELOG and documentation updates. Full coalesce-vue suite passes (683 tests).