voice: use 2x regular release for fast release

[sapphire-arches]

This is much more conservative than the original logic, avoiding note-off clicks for soft culls.

[github-actions]

Test Results

36 tests  ±0   36 ✅ ±0   0s ⏱️ ±0s
 7 suites ±0    0 💤 ±0 
 7 files   ±0    0 ❌ ±0 

Results for commit 217aa7a. ± Comparison against base commit d5ab4ee.

♻️ This comment has been updated with latest results.

[nikodemus]

i gotta be honest i have no idea why this works, but it helps with the House 15 song from chewabledrapery.

I'm not sure if this goes against project policy, but I at least would appreciate linking to the song in question. (Maybe also mentioning the reason/uncertainty of this in a comment, so one doesn't need to trawl commit messages to figure out why this is the way it is?)

[sapphire-arches]

i gotta be honest i have no idea why this works, but it helps with the House 15 song from chewabledrapery.

I'm not sure if this goes against project policy, but I at least would appreciate linking to the song in question. (Maybe also mentioning the reason/uncertainty of this in a comment, so one doesn't need to trawl commit messages to figure out why this is the way it is?)

yeah, i asked chewabledrapery to file a bug and it looks like they have, which has the song -- this PR is currently underbaked since i just wanted to get something for them to test today. I'll figure out why this works and retarget it to community before marking it ready for review

[sapphire-arches]

Cleaned up the commit message and thought the logic through correctly, should be better now.

[m-m-adams]

I believe this is actually making fast release at least twice as fast as normal release

A release of 0 has an increment of 8192 iirc. A fast release from polyphony limits (mono/auto/legato) is at an increment of 4096, which is doubled by this PR. A soft cull is at an increment of 65536, and so won't be affected

Good to merge though since it does seem like a good change