[WASM] Some examples don't work on Android WASM

[Christywl]

Test Env:
Chromium Version: nightly build 70.0.3503.0 (8755e6b)
Platform: Android 8.1.0(Google Pixel 2)

Expected Result:
The examples work on Android WASM.

Actual Result:
The following examples don't work on Android WASM:

• Inception V4(TFlite)
• Incep. Res. V2(TFlite)
• ResNet50 v2(Onnx)
• DenseNet(Onnx)
  
  

How to Reproduce:
1.Get the latest codes(1952226): git clone https://github.com/intel/webml-polyfill, and run npm install and npm run build
2. Download the example models according to https://github.com/intel/webml-polyfill/tree/master/examples/image_classification/model
3. Run: npm start
4. Visit http://127.0.0.1:8080/examples/image_classification/index.html
5. Switch to WASM backend and select the example:

• Inception V4(TFlite)
• Incep. Res. V2(TFlite)
• ResNet50 v2(Onnx)
• DenseNet(Onnx)

[Wenzhao-Xiang]

After build WASM with -s ASSERTIONS=2, found these two errors come from one same problem that the model is too large and allocating memory fails when initializing.
256M memory is not enough for these four models. Therefore, wasm will use Memory.grow() to allocate another 256M memory(Memory * 2) for use.
Seems sometimes this allocating operation will fail in Android.
That's the root cause.

After restart chrome, these models work again.

[Christywl]

This issue can't reproduce with webml-polyfill commit 466f88a.

[pinzhenx]

These four models still failed in WASM on Android 9

[Christywl]

This issue still happens for example and benchmark on Android, so reopen it.

[liyananxl]

Samsung S9(Android 8, Chrome Stable Version: 76.0.3809.89) can reproduce this bug on Inception V4(TFlite) and ResNet50 v2(Onnx).