use c implementation of quantization

scripts/all_reduce_test.py

@@ -51,7 +51,7 @@ def main(config: Config):
             globals={"all_reduce": all_reduce, "mat": mat, "transfer_dtype": transfer_dtype},
         )
     elif config.transfer_dtype is not None and torch.uint8:
-        from zeroband.compression import uniform_8bit_quantize
+        from zeroband.C.compression import uniform_8bit_quantize
 
         t0 = benchmark.Timer(
             stmt="all_reduce(mat, quantization_func=foo)",