fix: pin keyword index prefix=False round-trip behavior over gRPC (#1356)

grpc.KeywordPrefixParams is an empty message: presence is the only
signal, so an explicit prefix=False cannot be represented in gRPC.
It is sent as absent (same server-side semantics, disabled) and is
recovered as None. Document this at both conversion sites and pin
the behavior with a reverse-direction (rest->grpc->rest) test.

qdrant_client/conversions/conversion.py

@@ -1689,6 +1689,8 @@ class GrpcToRest:
             on_disk=model.on_disk if model.HasField("on_disk") else None,
             memory=cls.convert_memory(model.memory) if model.HasField("memory") else None,
             enable_hnsw=model.enable_hnsw if model.HasField("enable_hnsw") else None,
+            # presence of grpc.KeywordPrefixParams is the only signal, an explicit `prefix=False`
+            # cannot be represented in grpc and comes back as `None`
             prefix=True if model.HasField("prefix") else None,
         )
 
@@ -4383,6 +4385,9 @@ class RestToGrpc:
             is_tenant=model.is_tenant,
             on_disk=model.on_disk,
             enable_hnsw=model.enable_hnsw,
+            # grpc.KeywordPrefixParams is an empty message, its presence enables prefix matching,
+            # so an explicit `prefix=False` is sent as absent (which the server also treats as
+            # disabled) and is recovered as `None` when converting back
             prefix=grpc.KeywordPrefixParams() if model.prefix else None,
             memory=cls.convert_memory(model.memory) if model.memory is not None else None,
         )

tests/conversions/test_validate_conversions.py

@@ -643,3 +643,30 @@ def test_convert_points_update_operation_falsy_shard_key():
                 "shard_key_selector"
             ), f"{type(operation).__name__} dropped shard_key={shard_key!r}"
             assert GrpcToRest.convert_shard_key_selector(inner.shard_key_selector) == shard_key
+
+
+def test_convert_keyword_index_params_prefix():
+    from qdrant_client import models
+    from qdrant_client.conversions.conversion import GrpcToRest, RestToGrpc
+
+    def round_trip(prefix):
+        rest_params = models.KeywordIndexParams(
+            type=models.KeywordIndexType.KEYWORD, prefix=prefix
+        )
+        grpc_params = RestToGrpc.convert_keyword_index_params(rest_params)
+        return grpc_params, GrpcToRest.convert_keyword_index_params(grpc_params)
+
+    grpc_params, recovered = round_trip(True)
+    assert grpc_params.HasField("prefix")
+    assert recovered.prefix is True
+
+    grpc_params, recovered = round_trip(None)
+    assert not grpc_params.HasField("prefix")
+    assert recovered.prefix is None
+
+    # grpc.KeywordPrefixParams is an empty message whose presence enables prefix matching,
+    # so an explicit `prefix=False` cannot be represented in grpc: it is sent as absent
+    # (which the server also treats as disabled) and is recovered as `None`
+    grpc_params, recovered = round_trip(False)
+    assert not grpc_params.HasField("prefix")
+    assert recovered.prefix is None