## Problem

When Vertex AI returns a streaming response containing a JSON array split across multiple SSE chunks, the response parser does not accumulate the partial JSON across chunks. Each chunk is parsed independently, causing `json.JSONDecodeError` on partial array fragments and dropping content from all but the last chunk that completes the array.

## Proposed fix

Maintain a string buffer per stream that accumulates incoming JSON list content across chunks. Flush the buffer and parse when a complete JSON array is detected. This mirrors the existing partial-object accumulation logic and brings list handling to parity.

## Files

- `litellm/llms/vertex_ai/` (streaming response parser)