fix(#40): ensure RepeatedField<T>.AsArray() produces an array of exac…

…t length

- Re-allocate to a new exactly-sized T[] if the underlying buffer length of RepeatedField<T> does not match the count
- Bump up the submodule commit (no functional changes)
- Update TODO

TODO.md

@@ -23,6 +23,7 @@
 
 ## Nice to have
 
+- [ ] It appears the behavior of the gRPC stack has changed and inner `RepeatedField<T>` buffer is no longer created to the exact length of data received in the response. Analyze the implementation changes and consider options to reduce copying and allocations.
 - [ ] Add an extension to create an index and then wait for it to be ready to use, maybe add guards before sending requests?
 - [ ] Restructure the solution and add Pinecone.SemanticKernel project for long-term memory integration? Or is it better to directly contribute it to SK?
 - [ ] Consider putting the handling of "MetadataValue as DU" into a dedicated place instead of switches in multiple places. First-class DUs in C# when?

src/Grpc/Converters.cs

@@ -146,7 +146,13 @@ public static RepeatedField<T> AsRepeatedField<T>(this T[] source) where T : unm
 
     public static T[] AsArray<T>(this RepeatedField<T> source) where T : unmanaged
     {
-        return (T[])FieldAccessors<T>.ArrayField.GetValue(source)!;
+        var buffer = (T[])FieldAccessors<T>.ArrayField.GetValue(source)!;
+        if (buffer.Length != source.Count)
+        {
+            buffer = buffer[..source.Count];
+        }
+
+        return buffer;
     }
 
     public static void OverwriteWith<T>(this RepeatedField<T> target, T[]? source) where T : unmanaged