Improve reverb predelay to delay wet sound

[ueshita]

This PR modifies the Reverb predelay to separate dry from wet,
which is common in other software reverbs.

Closes: #98863

Why should dry and wet be separated?

Unlike direct sound, reverberation reaches our ears later.
Separating dry and wet preserves the clarity of dry, and we perceive the scale of the room.

The following captured audio were processed with the default parameter AudioEffectReverb for the footstep SFX.

before.mp4

after.mp4

[fkeyzuwu]

After some digging, the way to solve the problem seems pretty messy.
I believe this PR is the wrong approach.

In short, there's a few problematic things, but they all stem from one thing:
I am pretty sure the parameter named predelay isn't supposed to be named predelay like in traditional reverb effects. What I believe it does is specify the time between the early reflections, and feedback is how much early reflections there are. And in that sense it works correctly, just named incorrectly. The docs also don't say this.

Fixing this without breaking behavior compatibility for any project using the reverb would be problematic, since the current parameters are responsible for some part of the reverb effect, so we would want to preserve them. Changing the naming of these to fit with what they are actually doing and then adding an additional predelay parameter would break API compatibility. Changing the behavior of the predelay and repurposing the old behavior of the early reflections to new parameters would change the behavior of the effect in existing projects.

The only real way I can think of that wouldn't break compatibility is to add a new parameter, which will be the "predelay" we know from normal reverbs, but naming it something else. However this would add even MORE confusion. This will also be a new feature and not a bugfix at that point.

What we can maybe do to not break anything or make anything more confusing is to change the documentation of these parameters. We can acknowledge these problems and tell users that these parameters mean something different than what the parameter says, and is kept that way in order to avoid breaking changes, while keeping the reverb DSP code the same, and not adding in predelay in the traditional sense.

Don't really know what is the best approach in this situation.

[ueshita]

Thank you for your comment.

In short, there's a few problematic things, but they all stem from one thing:
I am pretty sure the parameter named predelay isn't supposed to be named predelay like in traditional reverb effects. What I believe it does is specify the time between the early reflections, and feedback is how much early reflections there are. And in that sense it works correctly, just named incorrectly. The docs also don't say this.

I agree with this.
In the dsp code, it's an echo_buffer and I think this is the behavior of the early reflection.
It would be difficult to maintain compatibility with this modification.

The only real way I can think of that wouldn't break compatibility is to add a new parameter, which will be the "predelay" we know from normal reverbs, but naming it something else. However this would add even MORE confusion. This will also be a new feature and not a bugfix at that point.

For example, the new parameter could be initial_delay, which might be less confusing.
(This is something we should talk about in godot-proposals.)