fix(partial): don't fail final stream validation on explicit nulls for required nullable fields

[sohumt123]

Describe your changes

Streaming with create_partial raises a spurious ValidationError on valid, completed streams whenever the response model has a required-but-nullable field (e.g. email: Optional[str] with no default) that the LLM legitimately returns as JSON null.

Repro (fails on main at 47fdb2c, both sync and async):

from typing import Optional
from pydantic import BaseModel
from instructor.dsl.partial import Partial

class User(BaseModel):
    name: str
    email: Optional[str]  # required, but nullable

chunks = ['{"name": "Al', 'ice", "email"', ': null}']
for obj in Partial[User].model_from_chunks(chunks):
    print(obj)
# pydantic_core ValidationError: 1 validation error for User
# email: Field required [type=missing, input_value={'name': 'Alice'}, ...]

Root cause: after the last chunk, PartialBase.model_from_chunks / model_from_chunks_async re-validate the accumulated object via

original_model.model_validate(final_obj.model_dump(exclude_none=True), **kwargs)

The model_dump(exclude_none=True) round-trip cannot distinguish "field explicitly null in the JSON" from "field never streamed" (both end up as None on the constructed partial), so explicit nulls are stripped from the dump and pydantic reports the field as missing — even though the stream produced complete, valid JSON and the per-chunk validation of the complete root already succeeded. The same round-trip would also mis-validate models with Field(exclude=True) fields.

Fix: validate the accumulated JSON string itself (parsed with jiter.from_json, already imported) instead of the lossy dump, in both the sync and async paths. This preserves explicit nulls while keeping the guard's intent intact:

• a required field genuinely absent from the completed JSON still raises (existing TestFinalValidationAfterStreaming tests still pass), and
• structurally incomplete streams still skip final validation.

Tests: added three regression tests in tests/dsl/test_partial.py — explicit-null nullable field completes cleanly (sync + async; both fail on main with the exact error above), and a genuinely-absent nullable field still raises. Ran pytest tests/dsl/ tests/coverage/test_dsl_partial_coverage.py: 90 passed, 2 skipped. ruff check/ruff format --check and ty check clean on the touched files. (The pre-existing test_citation_keeps_quotes_without_context_and_recovers_fuzzy_quotes failure in tests/coverage/test_dsl_small_coverage.py reproduces identically on a clean checkout and is unrelated — it's targeted by #2430/#2443.)

Issue ticket number and link

None — found while exercising partial streaming with nullable required fields.

Checklist before requesting a review

• I have performed a self-review of my code
• If it is a core feature, I have added thorough tests.
• If it is a core feature, I have added documentation.