Fix missing tree_reduce import in models/cache.py

[siiea-ai]

_BaseCache.nbytes calls tree_reduce, but the module only imports tree_flatten, tree_map, tree_unflatten from mlx.utils. Reading .nbytes on any cache that inherits the base implementation (hit by MoE architectures like Qwen3.5 MoE) raises NameError: name 'tree_reduce' is not defined.

Fixes #1164.

Diff

-from mlx.utils import tree_flatten, tree_map, tree_unflatten
+from mlx.utils import tree_flatten, tree_map, tree_reduce, tree_unflatten

Repro (before the fix)

curl -sS -X POST http://127.0.0.1:8888/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"qwen3.5-397b-a17b-mlx","stream":false,"max_tokens":20,
       "messages":[{"role":"user","content":"Say hello."}]}'
# → {"error":"Error in iterating prediction stream: NameError: name 'tree_reduce' is not defined"}

After the patch, the same request returns a normal completion. Verified on LM Studio 0.4.12+1 with bundled mlx-lm 0.31.3 on macOS Apple Silicon.

[angeloskath]

Thanks!