### What happened?

Every **streaming** call to the `amazon_nova` provider fails. Non-streaming on the same account, model and prompt succeeds.

``` litellm.MidStreamFallbackError └─ httpx.DecodingError: cannot use a decompressobj multiple times └─ zstandard.backend_c.ZstdError: cannot use a decompressobj multiple times ```

Reproduced **5/5** streaming attempts against `amazon_nova/nova-micro-v1`.

### Root cause

`https://api.nova.amazon.com/v1` returns `content-encoding: zstd` for streaming SSE when the client offers it, and emits the stream as **independent zstd frames**. Verified with `curl` — a 785-byte capture contains 5 frame magics (`28b52ffd`) at offsets `[0, 210, 405, 584, 762]`.

`httpx.ZStandardDecoder.decode()` resets its decompressor only when `unused_data` is non-empty *within the same call*:

```python output.write(self.decompressor.decompress(data)) while self.decompressor.eof and self.decompressor.unused_data: ... # only resets if leftover bytes are present in THIS call ```

One SSE flush per network chunk delivers exactly one complete frame, so the decoder ends at `eof=True` with empty `unused_data`, and the next `decode()` reuses a finished decompressob…