fix(python): accept none dimension in VectorSchema

[HosniBelfeki]

What

VectorSchema declares dimension: Optional[int] = 0, and its docstring states
the value "may be None for sparse vectors"
(python/zvec/model/schema/field_schema.py:204-205, 225).

The constructor rejects it. isinstance(None, int) is False, so line 241 raises:

>>> VectorSchema("sparse", DataType.SPARSE_VECTOR_FP32, dimension=None)
ValueError: Invalid schema: vector's dimension must be >= 0

Two problems: the documented and annotated None does not work, and the error
reports a range violation for what is actually a type rejection.

Severity

Low, and not a security issue. Nothing is silently wrong and no data is
affected — a user who follows the documented sparse-vector idiom just hits a
confusing error. The repo's own fixture sidesteps it by omitting the argument
(python/tests/detail/fixture_helper.py:78), which is likely why it went unnoticed.

Fix

Normalize None to 0 before the range check. This mirrors what the same
constructor already does for its other Optional parameter, nine lines below:

Parameter	Declared	None handling
index_param	Optional[...] = None	normalized to FlatIndexParam() (line 251)
dimension (before)	Optional[int] = 0	raises ValueError
dimension (after)	Optional[int] = 0	normalized to 0

0 is the value sparse vector fields already carry, so this changes no existing
behavior — inputs that worked before produce identical schemas. Dense fields
declared with dimension 0 are still rejected downstream by the C++ validator
(field[...]'s dimension must be in (0,20000]), which is where the check belongs,
since 0 is legal for sparse.

Tests

• test_dimension_none pins the documented contract. Without the fix it fails with
  ValueError: Invalid schema: vector's dimension must be >= 0; with it, the schema
  builds and dimension == 0.
• test_dimension_invalid guards what the fix touches — -1 and "128" must still
  be rejected. It passes both with and without the change.

Verified on Linux (Debian 13 container, x86_64, CPython 3.12.14) and on Windows 11
(CPython 3.12.4). pytest python/tests/test_schema.py → 11 passed. Reverting the
three-line change while keeping the tests gives 1 failed, 10 passed on both
platforms, failing with the ValueError quoted above.

ruff==0.14.4 (CI's pin): ruff check . → All checks passed!,
ruff format --check → 2 files already formatted.