In bug 1521370 we saw a great number of crashes due to a problem when loading the WMF libraries.

The solution was to add a crashguard around the creation of the WMF decoder. However, it still causes the content process to crash at least once; additionally such a crash is highly noticeable as it requires the user to reload the tab.

A WMF decoder is always created on the content process when querying the GPU decoder if a particular codec is supported.

We shouldn't, instead we should query the GPU process on what codec it supports over an IPC call.

No WMF calls should be done in the content process.

This was done in bug 1595994

Hardware WMF decoding will occur in the GPU decoder and software will happen in the rdd process.

We will fall-back to in content if neither are supported.

We could remove that fall-back.

The advantage of this in that we will only load the WMF framework once in each process.