feat(build): bazel build support flatbuffer serialization

GitOrigin-RevId: 4755c27e06

.gitattributes

@@ -18,3 +18,4 @@ ci/resource/dump/roi_align_backward_8.8.0.mdl filter=lfs diff=lfs merge=lfs -tex
 ci/resource/dump/relayout_format_8.10.0.mdl filter=lfs diff=lfs merge=lfs -text
 ci/resource/dump/batch_conv_bias_with_policy_8.8.0.mdl filter=lfs diff=lfs merge=lfs -text
 ci/resource/prof/model_with_err_assert.mdl filter=lfs diff=lfs merge=lfs -text
+ci/resource/prof/test_mge.mge filter=lfs diff=lfs merge=lfs -text

src/serialization/impl/flatbuffers_helper.cpp

@@ -67,6 +67,9 @@ megdnn::DType load_dtype(const fbs::DType* dtype) {
             return dtype::Quantized4Asymm{param->scale(), param->zero_point()};
         case DTypeEnum::DTypeEnum_Quantized8Asymm:
             return dtype::Quantized8Asymm{param->scale(), param->zero_point()};
+        default:
+            // Float16 may be disabled
+            megdnn_trap();
     }
     return {};
 }

src/serialization/impl/serializer_oss.cpp

@@ -638,12 +638,12 @@ void GraphLoaderOSS::OprLoadContextImpl::load_tensor_value(
     mgb_throw_if(
             consumed_size > data_size, SerializationError,
             "Custom tensor value loader consumed more data than "
-            "available: consumed %lu, has %u",
+            "available: consumed %zu, has %u",
             consumed_size, data_size);
     if (consumed_size < data_size) {
         mgb_log_warn(
                 "Tensor value loader consumed less data than available: "
-                "consumed %lu bytes, has %u bytes",
+                "consumed %zu bytes, has %u bytes",
                 consumed_size, data_size);
         file->skip(data_size - consumed_size);
     }