InMemoryVectorStore Silently Drops Documents When Embedding Returns Mismatched Vector Count

The `add_texts` and `aadd_texts` methods of `InMemoryVectorStore` use `zip(texts, vectors)` without `strict=True`. Python's `zip` defaults to `strict=False`, which silently truncates to the shortest iterable. When `Embeddings.embed_documents` returns fewer vectors than the number of input texts (e.g., because of an off-by-one bug in a custom embedding implementation or a batched provider truncation), the extra documents are never stored, and the caller receives fewer IDs with no error.

1. Define a custom `Embeddings` subclass that returns one fewer vector than the number of input texts in `embed_documents`. 2. Instantiate `InMemoryVectorStore` with that embedding. 3. Call `store.add_documents([Document('first'), Document('second'), Document('third')])`. 4. Observe that it returns only two IDs and no exception is raised, while the third document is silently dropped.

This fix requires Python 3.10 or later because `zip(..., strict=True)` was introduced in Python 3.10. If the project still supports Python 3.9 or earlier, the `strict=True` argument will raise `TypeError`; in that case, use only the explicit length check and omit `strict=True`. The change may cause previously silent embedding truncations to surface as runtime `ValueError`s, which is intended but could break existing deployments that accidentally relied on the old behavior. InMemoryVectorStore is not thread-safe for concurrent writes, and this fix does not change that; concurrent callers may still lose data if the same store is modified without external synchronization. Rollback is possible by removing the length checks and `strict=True`, but doing so will re-introduce silent data loss.