Add example on how to replace Conv2DTranspose with UpSampling2D

FAQ.md

@@ -447,8 +447,20 @@ Why are `Conv2DTranspose` layers not supported?
 `UpSampling2D` layers seem to be the better alternative:
 https://distill.pub/2016/deconv-checkerboard/
 
+Basically, instead of this:
+
+```python
+x = Conv2DTranspose(8, (3, 3), strides=(2, 2), padding='same')(x)
+```
+
+one uses that:
+
+```python
+x = Conv2D(8, (3, 3), padding='same')(UpSampling2D(2)(x))
+```
+
 In case you are not in the position to change your model's
-architecture to use `UpSampling2D` instead of `Conv2DTranspose`,
+architecture to make that change,
 feel free to implement `Conv2DTranspose` in frugally-deep and
 submit a [pull request](https://github.com/Dobiasd/frugally-deep/pulls). :)