v6: JSON codec — records <-> bytes, type handling

[Pandys]

_codec.py: the data-plane encoding used by all DataFrame flows. Deliberately pandas-independent (records in/records out) so Arrow/polars surfaces stay additive.

• records -> JSON bytes for InsertRequest.data / update / delete payloads; chunked (~2500 rows default, matching 5.x behaviour)
• AggregateResponse.data JSON bytes -> records
• NaN/NaT -> JSON null; timestamps -> ISO-8601; document _id handling (kept as regular column for round-trips)
• Pipeline encoding follows the API rules: BSON-encoded Mongo pipeline for Mongo-backed collections, JSON-encoded Elastic DSL for Elastic-backed (server picks backend from the collection's storage descriptor)
• Mongo Extended JSON v2 forms (ObjectID, Date) parsed to sensible Python types