Fix further loop filtering issues

[tristanphease]

From #148 a few images were still failing to match the libwebp output precisely so investigated and found 2 further issues:

1. From the spec: https://datatracker.ietf.org/doc/html/rfc6386#section-15
   Note carefully that loop filtering must be skipped entirely if loop_filter_level at either the frame header level or macroblock override level is 0. In no case should the loop filter be run with a value of 0; it should instead be skipped.
   We are currently only checking if the final loop filter value is set to 0 then we skip loop filtering but if the frame has loop filter level set to 0 we should skip altogether. Therefore just return early if we detect that.
2. We're not actually using the loop filter adjustments decoded in the frame header. The spec is a bit confusing here so just tried to match libwebp where if it's enabled we just add the first value. https://github.com/webmproject/libwebp/blob/85e098e58d3fb89fee93b09c6c4515dbc4593c46/src/dec/frame_dec.c#L295C1-L300C10
   edit: checked the spec's implementation and now it makes sense, we only use the first value since it matches the first frame of the vp8

After these changes, those images that were not matching now do in my local tests, so change the decoder tests to match exactly.

[tristanphease]

Hey @Shnatsel would appreciate if you could run your tests again on all your images with this change, my method of checking the images isn't very clean.

[Shnatsel]

With this change decoding is bit-exact with dwebp -nofancy on the entire corpus 🎉