.Net: Python: Bug: SqlServerCollection search fails with the default distance function and orders dot results farthest-first #14713

Describe the bug

SQL Server vector searches are ordered through the shared DISTANCE_FUNCTION_DIRECTION_HELPER, which treats dot product as a similarity. Two user-visible failures follow:

• A vector field left at its default distance function (VectorStoreField("vector", dimensions=...) → DistanceFunction.DEFAULT) is accepted by DISTANCE_FUNCTION_MAP (mapped to cosine) but has no entry in the direction helper, so every SQL Server search fails at runtime with KeyError: <DistanceFunction.DEFAULT: 'DEFAULT'>.
• A field using DistanceFunction.DOT_PROD builds ORDER BY ... DESC, returning the least similar rows first.

VECTOR_DISTANCE is a distance for every metric SQL Server supports; the dot metric is the negative dot product and Microsoft documents that smaller values are more similar.

To Reproduce

No database or model call is needed to construct the query:

from semantic_kernel.connectors.sql_server import _build_search_query
from semantic_kernel.data.vector import DistanceFunction, VectorSearchOptions, VectorStoreField

key = VectorStoreField("key", name="id", type="str")
vector = VectorStoreField("vector", name="embedding", type="float", dimensions=3)
print(vector.distance_function)  # DistanceFunction.DEFAULT

_build_search_query(
    "dbo",
    "test",
    key,
    [],
    [vector],
    [0.1, 0.2, 0.3],
    VectorSearchOptions(vector_property_name="embedding"),
)
# KeyError: <DistanceFunction.DEFAULT: 'DEFAULT'>

The same failure is reachable through SqlServerCollection.search(vector=...) when the model's vector field omits distance_function. With distance_function=DistanceFunction.DOT_PROD, the generated query ends with ORDER BY _vector_distance_value DESC.

Expected behavior

The default search should return results ordered by cosine distance, nearest first. Dot-product results should also be ordered ascending: SQL Server's VECTOR_DISTANCE('dot', ...) is the negative dot product, and smaller values indicate more similar vectors.

Screenshots

Not applicable.

Platform

• Language: Python
• Source: main at d41312f
• AI model: Not applicable (query construction only)
• IDE: Not applicable
• OS: Windows 11

Additional context

InMemoryCollection already resolves DEFAULT to COSINE_DISTANCE. The .NET SQL Server connector made the same dot-product ordering fix in #13071 (NegativeDotProductSimilarity from DESC to ASC). I have a fix and regression tests ready and will open a PR linking this issue.