SPU: Reintroduce simpler play delay

[Ziemas]

Description of Changes

Reintroduces a delay before starting to play keyed on voices, but in a simpler way than before.

Rationale behind Changes

Seems to be needed for Dragon Quest V's streamed music. It uses the SPU IRQ to track playback progress, but starts the voices before enabling it which means the voices will have already moved past its IRQ point and breaking its state machine.

Fixes #13486

Suggested Testing Steps

This needs testing in Battlefield 2: Modern Combat, with the vehicle sounds in online multiplayer. Idk if this can be tested offline?
I was able to test battlefield myelf, still works.

Did you use AI to help find, test, or implement this issue or feature?

No.

[Ziemas]

Hm, I tried testing some stuff on hardware and got even more confused.

The real SPU actually spends even less time in the first ADPCM block than pcsx2 without this patch.

I think I'm going to end up just adding the per-voice sample buffers so I can compare more easily.

The actually correct fix for this is probably to IRQ test the block header address for each voice on every tick.

[Ziemas]

Rebased on top of #13497.

Just for fun I thought I'd see if i could match the timings I see on hardware.

Here is an example of starting a voice on the PS2, after the voice is keyed on it seems like it takes about an audio frame to see a new NAX value. There is then a larger delay before the next one before decoding continues as normal.

ps2:

17.306348: [+8] Keyed on at [2806]
17.306375: [+27] NAX switched to new sample [4002]
17.306429: [+54] NAX switched to new sample [4003]
17.306439: [+10] NAX switched to new sample [4004]
17.306459: [+20] NAX switched to new sample [4005]

By only running the decoding step on odd numbers of the play delay we can see a very similar looking pattern.
pcsx2:

pcsx2
0.182873: [+4] Keyed on at [2803]
0.182899: [+26] NAX switched to new sample [4002]
0.182941: [+42] NAX switched to new sample [4003]
0.182962: [+21] NAX switched to new sample [4004]
0.182983: [+21] NAX switched to new sample [4005]

(the format of those logs is sec.usec: [usec delta from last event])

Idk if this is particularly worthwhile to implement though.

[Ziemas]

Not bothering with this for now unless we find something that needs it.