fix: buy each perishable on the first trip or the latest trip before use

[guplem]

Summary

Closes #77

The shopping planner now puts each perishable on the first trip (usually a home delivery) when it stays fresh from there. Otherwise it goes on the latest trip that keeps it fresh, so you buy it close to its use. Before, it went on the earliest trip that kept it fresh, which could be an in-between trip.

• Planner (multi_trip_planner.dart): the greedy pass still picks the trips. After the trip set is final, a new pass gives each perishable its trip with _tripForPerishable. Freeze-on-arrival items and the "no trip keeps it fresh" fallback do not change. Non-perishables now get the first trip after the trip set is final, so a later trip never stays open for a pantry item alone.
• One fewer trip in a rare case: when the first trip keeps every item of a later trip fresh, that later trip is now empty and disappears. Example: milk that keeps 30 days for day 10, plus salt for day 3, was 2 trips and is now 1. This follows the rule ("the delivery takes everything it can") and saves a trip. A test pins it.
• Docs: ADR 0014 records the rule and the reason; AGENTS.md test count is 1073; ADR 0014 and 0015 no longer claim the plan is always the minimum number of trips.
• The copied list, the PDF and the shopping page read the planner, so they follow without a change.

Test plan

• 7 new planner tests (yogurt to the latest trip, cheese to the first trip, dropped empty trip, pantry items with no extra trip (2), one-trip mode), each proven by a mutation; 1 fallback regression guard
• Test helper compares full trip items (amount, unit, cook days, freeze flag)
• No existing test changed
• Format check, flutter analyze, full suite (1073 tests) pass locally

🤖 Generated with Claude Code

[guplem]

The new rule is sound: every perishable lands inside its freshness window, lastWhere cannot throw because the greedy pass put a trip in every window, and the fallback and the maxWeekIndex cap behave as before. Consumers compare against trips.first.weekIndex, so a trip that disappears breaks nothing.

• [Required] A later trip can survive for a pantry item alone (multi_trip_planner.dart, non-perishable loop). Non-perishables get their trip before the trip set is final, in timeline key order. Timeline {milk: day 10 (keeps 30 days), salt: day 10, pepper: day 3}: milk takes trip 1, salt rides trip 1, pepper adds trip 0, then the new pass moves milk to trip 0. Trip 1 now holds only salt, so the user makes an extra trip for an item that keeps forever, and the result depends on map order. Fix: in the final pass, give every non-perishable chosenTrips.first, and add a test that expects one trip.
• [Suggestion] _itemsByTrip in the tests drops amount, unit and freezeOnArrival, so the new tests cannot catch a wrong amount or a lost freeze flag. Include them.
• [Suggestion] No test covers one-trip mode (assumeFreezerForFreezable: true), the main place where trip 0 appears after the greedy pass. Add one: freezable chicken for day 14 plus non-freezable cheese for day 10 (keeps 30 days) end on one trip.
• [Suggestion] ADR 0014 line 87 and ADR 0015 line 66 claim the plan is always the minimum number of trips. That holds for perishables only; a trip 0 added later can make the plan bigger. Limit the claim.
• [Nitpick] The planner doc comment (lines 49-61) and the new comments say "keeps it fresh" and "latest trip that still gets the item fresh". The code checks the [earliestWeek, latestWeek] window, and a fallback window can hold a trip that does not keep the item fresh. Describe the window.
• Issue #77 criterion 1 ("the trip set stays the same") is now reworded to allow a later trip to disappear.