Normalizer stereo imaging and performance improvements

[roderickvd]

Previously, the normalizer integrator and peak were recalculated as if the samples were a mono stream. If actually one channel required limiting and the other not, this could cause nervousness in the stereo imaging.

This PR changes the normalizer to:

• calculate the integrator and peak for each channel individually
• then couple the left/right channel by using the limit gain of each interleaved pair of samples
• be faster by optimizing the hot code path and reducing the number of branches

Again, please test this for me as I don't have a Spotify account anymore. I ported it from pleezer with my permission.
My "go to" track to test limiting is "Blumine" by Mahler on "Masterpieces in Miniature".

[Copilot]

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

Comments suppressed due to low confidence (2)

playback/src/player.rs:1611

• [nitpick] Consider renaming 'normalisation_channel' to something like 'current_channel' (or adding a clarifying comment) to better convey its purpose as a toggling index for stereo channels.

self.normalisation_channel ^= 1;

playback/src/player.rs:1758

• Review the change in the state match pattern: switching from 'PlayerState::Invalid { .. }' to 'PlayerState::Invalid' may bypass handling of extra data previously matched; ensure that this change correctly covers all intended invalid states.

if matches!(self.state, PlayerState::Invalid) {

[photovoltex]

So I used the state of the branch for a bit (with normalization enabled) and it seems to not have (at least for me) noticeable interference or audio spikes. I will test it for a day more and then we could probably get it merged :D

[photovoltex]

So, testing wise I would approve these changes. Just two small comments that didn't make quite a lot of sense to me.

[photovoltex]

Ah dang it, forgot to approve it sooner. After the conflict is resolved you can merge it :)

[roderickvd]

No prob I'll make some minor changes anyway.