Cleanup `CudaDeviceInterface::getCudaContext()`

[scotts]

The implementation of CudaDeviceInterface::getCudaContext() is unnecessarily confusing:

1. There is a #if on the FFmpeg version directly inside the implementation to dispatch to different functions.
2. Outside of the function, we #if on the FFmpeg version to define two different functions.
3. The only difference between both functions is one parameter to av_hwdevice_ctx_create(), but all of the surrounding code is the same.

What we should do:

1. Create a single function in FFMPEGCommon that internally does a #if on FFmpeg version, calling av_hwdevice_ctx_create() in the appropriate way for each version.
2. Remove getFFMPEGContextFromExistingCudaContext(), getFFMPEGContextFromNewCudaContext() and the #if surrounding them.
3. Call the function defined in 1 directly in getCudaContext() with the logic removed in 2 directly in getCudaContext().

There's two main principles we want to apply here:

1. Keep FFmpeg version #if out of decoding code as much as possible. We want to abstract it into a function in FFMPEGCommon that decoding code can call.
2. The minimum code necessary should be inside the FFmpeg version #ifs.

[NicolasHug]

Let's also change the name of getCudaContext() into something more accurate: it's returning a "hardware device context", which is an FFmpeg concept. It's not a CUDA context, which also exists, but is a different concept entirely.

Similarly:

• we should avoid calling this hardware device context an "FFmpeg context", like it's currently named in getFFMPEGContextFromExistingCudaContext (which we want to remove). There's no such thing as a global FFmpeg context so let's stick to "hardware device context".
• Let's rename the ctx_ field of the interface to reflect that it's a hardware device context - we have so many contexts around, it's worth having more descriptive names so we can easily distinguish objects.