negative_insulin: add version compatible with Loop next-dev

[bjorkert]

The existing negative_insulin patches do not apply to the Loop next-dev branch, so the customization shows as incompatible.

On next-dev the glucose-prediction pipeline (including DoseMath and the insulin model) moved into the standalone LoopAlgorithm Swift package, which no longer depends on LoopKit. The original approach — a doseEnd parameter on DoseStore.getGlucoseEffects, the LoopKit insulin-model changes, LoopState.negativeInsulinDamper, etc. — no longer applies. nextdev_negative_insulin.patch keeps the same behavior with a smaller, retargeted change: the damper coefficient is computed in LoopDataManager and threaded into the algorithm as an optional AlgorithmInput.negativeInsulinDamper, with the damping applied inside LoopAlgorithm.generatePrediction.

The Negative Insulin Damper is an opt-in Algorithm Experiment, off by default. With it off the coefficient is nil and the algorithm behaves exactly as stock.

Test

Requires LoopWorkspace on the Loop next-dev branch. No lnl-scripts change is needed; only PATCH_BRANCH is overridden.

reset command

This command resets the branch to default state.

git submodule foreach 'git reset --hard; git clean -fd;'

apply

export PATCH_BRANCH="nextdev_negative_insulin"
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/loopandlearn/lnl-scripts/main/CustomizationSelect.sh)" \
    negative_insulin
unset PATCH_BRANCH

[marionbarker]

Updated code

The location in Loop/Loop.xcodeproj/project.pbxproj where Preferences.swift line is inserted for basal_lock and where NegativeInsulinDamperSelectionView.swift line is inserted for negative_insulin were too close together.

The lines in the negative_insulin/nextdev_negative_insulin.patch file were modified to shift the insertion of NegativeInsulinDamperSelectionView.swift a few lines earlier in the pxbproj file and reduce the number of comparison lines to avoid the location where basal_lock makes a change.

The basal_lock/nextdev_basal_lock.patch was similarly modified to reducs the number of comparison lines to avoid the location where negative_insulin makes a change.

Test

The application and removal of basal_lock and negative_insulin are now independent.

[marionbarker]

Test Summary

✅ The negative_insulin damper customization behaves as expected.

For this example glucose pattern, the Loop algorithm (black curve) provides more IOB when recovering from a low than when Negative Insulin Damper is enabled (pink curve).

The experiment was run using the method described in the Negative Insulin Damper report:

• Report for Negative Insulin Damper

[marionbarker]

Approved by code review and test. Added a few tweaks to deconflict basal lock and negative insulin damper.

Request @bjorkert review this again.

[bjorkert]

Approved, LGTM