Add ManagedObjectCache

[colemancda]

Fix quadratic performance in NSManagedObjectContext.insert(_:model:) batch inserts

Problem

Batch inserts degrade quadratically with batch size. Inserting a real-world dataset
(11,421 objects with relationships) took ~46 seconds, while the same data inserted
as 360 small groups took ~2.5s — the opposite of what batching should do.

Two causes, both in the find-or-create path:

1. insert(_ values:model:) issues one fetch request per object to find-or-create it.
   Core Data fetches evaluate their predicate against every pending (unsaved) object
   in the context, so as a batch accumulates inserted-but-unsaved objects, each
   subsequent find gets slower — O(n²) overall.
2. setRelationship resolves every relationship target with another per-object
   context.find, which pays the same pending-changes penalty.

Fix

insert(_ values:model:) now:

• Prefetches every object referenced by the batch — value IDs plus all relationship
  target IDs — with a single IN fetch request per entity, into a
  ManagedObjectCache ([EntityName: [ObjectID: NSManagedObject]]), before anything
  is inserted (pending set is empty, so these fetches are clean SQL).
• Creates without fetching: the prefetch covers every value ID, so a cache miss
  proves the object doesn't exist — no fallback fetch. (This mattered: with the
  fallback in place, the empty-database case still spent ~29s in fetches.)
• Creates all batch objects before applying values, so relationships between
  values in the same batch resolve regardless of their order in the array.
• Applies attributes/relationships through cache-aware setValues/setRelationship
  overloads, then saves once.

The single-value insert(_ value:model:) path is unchanged; the original
setValues/setRelationship signatures delegate through an empty cache.

Results

360-site catalog, 11,421 objects with to-many relationships (M-series Mac):

Scenario	Before	After
Single batch (1 save)	46.6s	0.26s (~175x)
360 groups of ~32 objects	2.5s	0.9s

Post-fix single-batch breakdown: prefetch 0.03s · create 0.02s · setValues 0.08s · save 0.12s.

Behavior notes

• Duplicate (entity, id) pairs in one batch resolve to the same managed object with
  last-write-wins, matching the previous loop's behavior.
• Intra-batch relationship resolution is now order-independent (previously a toOne
  referencing a value later in the array could resolve to nil).
• An unknown entity name now throws CoreModelError.invalidEntity up front instead of
  failing inside a fetch request.

Testing

• All existing tests pass.