[ps2]

Part of decoupling LoopKit and LoopAlgorithm (so Trio can use LoopKit without the algorithm). It depends on:

• Take plain numbers in fixed units at the public API LoopAlgorithm#17: the public API takes plain numbers in fixed units, with unit-suffixed names.
• Own the shared value types instead of depending on LoopAlgorithm LoopKit#674: LoopKit owns the shared value types.

With those, Loop is the one place that converts between LoopKit's types and the algorithm's.

What changes

• Loop/Models/AlgorithmBridge.swift:
  • StoredGlucoseSample and StoredCarbEntry conform to the algorithm's protocols.
  • LoopKit insulin models are wrapped for the algorithm.
  • Schedules convert to mg/dL; ISF, targets and the suspend threshold go in as sensitivityMgdlPerUnit, targetMgdl and suspendThresholdMgdl.
  • Algorithm results convert back to LoopKit values for storage, display, the watch and the extension context.
• Common/LoopKitTypeAliases.swift: makes unqualified names like GlucoseEffect mean LoopKit's in each target. Algorithm types are written LoopAlgorithm.X. The struct is now Algorithm, so calls are Algorithm.run(input:).
• Where LoopAlgorithm is linked: once, in LoopCore. The widget, watch app and LoopCore's own code no longer import it; 15 files in the app still do.
• LoopCore constants: gains the two constants the UI used from LoopAlgorithm (input recency, maximum carb absorption time).
• Current Setup: names the slow retrospective correction experiment that the LoopAlgorithm update brings in (Guard + Runway, second version; slow retrospective correction experiment (off by default) LoopAlgorithm#16). Loop doesn't turn it on.

Testing

• New end-to-end mmol/L test (testMmolSettingsDoseTheSameAsMgdl). With ISF, targets, suspend threshold and glucose all in mmol/L, the algorithm receives mg/dL values and Loop doses exactly as in the mg/dL test.
• Moved tests: two override tests moved from LoopKit; they check LoopKit's override schedules flowing into the algorithm. The dose annotation and IOB tests went to Take plain numbers in fixed units at the public API LoopAlgorithm#17 instead.
• LoopTests: 415 pass. The full LoopWorkspace build has all 19 plugins.
• Simulator run, on the prototype of this change before rebasing: closed loop with mock CGM and pump (automatic boluses, temp basals, carbs, the bolus screen). I saved the input of 17 algorithm runs and replayed each through the previous LoopAlgorithm. Forecasts, effects, IOB/COB and recommendations matched the app's output within 1e-9.