Simplify fast scenechange code a bit

[FreezyLemon]

downscaled_frame_buffer is an Option<([Plane<T>; 2], bool)>. The bool is called is_initialized and is supposed to show whether the memory in the [Plane<T>; 2] is initialized or not.

As far as I can tell, there is no code path that results in the Planes being uninitialized: downscaled_frame_buffer starts out as a None, the first buffer access allocates & initializes the memory, and the memory should always be initialized after that. The bool never seems to be false from what I can tell, which would support my assumption, I think.

There's also a field called frame_ref_buffer which caches the Arc<Frame<T>> when no downscaling is being done. Unless I am missing something, there's no real benefit to caching these pointers (because refs to both frames will always be passed into the function anyways), so i removed the buffer.

I tested my changes to the best of my ability, and didn't notice any change in behaviour or performance. LMK what you think

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (3374dc9) 88.48% compared to head (5f127b6) 88.48%.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #3320      +/-   ##
==========================================
- Coverage   88.48%   88.48%   -0.01%     
==========================================
  Files          88       88              
  Lines       28236    28235       -1     
==========================================
- Hits        24986    24985       -1     
  Misses       3250     3250              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[barrbrain]

@shssoichiro What are the implications for https://github.com/rust-av/av-scenechange/ ?

[shssoichiro]

It looks like this shouldn't break the scene detection code in either av-scenechange or av1an, as far as I can tell.

[shssoichiro]

As for why the code is like this, from what I recall at one point in time I believed there to be cases where we may be re-running a comparison on the same pair of frames twice. However, it's possible that is no longer true (or never was true) if there is no measurable difference in performance. This shouldn't affect behavior either way, as it was intended to be a performance optimization, but I'm okay with removing it.