Add the Ingredients and Areas tabs

[solcott]

Adds the Ingredients and Areas Home tabs end to end, then a follow-up pass that
fixes what a review against main turned up.

The feature

AreaEntity/AreaDao, an AreasFetchHistory DataStore, AreaApi on
list.php?a=list, AreaRepository, presenters and producers for both tabs, and
two grid screens. Ingredient fetch history collapses to a single timestamp,
matching the fact that its fetcher ignores the key and always pulls the whole
list. The Ktor logger moves to commonMain now that Kermit 2.2.0 no longer
needs the per-platform split.

Fixes on top

Browse-by-area fetched twice. recipesByArea drove the recipe Store's key
off areaRepository.getArea(). On a cold areas cache that flow emits
Data(null) first, producing ByArea(area, null) and starting a fetch, then
Data(Area), producing ByArea(area, "United Kingdom") — which flatMapLatest
uses to cancel the first stream and fetch again. Two round trips, a loading
flicker, and two entries in the fetch-history for one screen. country comes
back out of the key and is looked up once, suspending, inside the fetcher.
Verified against the running app: one filter.php?a=British and one
filter.php?a=United+Kingdom, issued together, no second round.

The two filter.php calls ran sequentially, and since the client sets
expectSuccess, a 404 on the country request discarded an area request that had
already succeeded. Now concurrent, with the country call best-effort and skipped
when it is the same word as the area.

strCountry is nullable in the v2 schema and NetworkJson has no
coerceInputValues, so one null row would have failed the decode of all 195.
Every row carries one today — this is hardening, not a fix for an observed
failure.

The areas writer upserted without evicting, so an area dropped upstream would
have stayed in the grid forever.

ktfmt never covered androidMain

Worth a look on its own. ktfmt-gradle 0.27.0 registers no task for any
androidMain source set in these modules: its KMP handler skips android*
source sets unless com.android.kotlin.multiplatform.library is applied, but it
runs from plugins.withId("org.jetbrains.kotlin.multiplatform"), which fires one
line before kmp.library applies that plugin; the Android handler it defers to
then returns early on an extension type it does not recognise. Two of the three
files it had let drift predate this branch. The formatting convention plugin
now registers the missing tasks, filtered by ktfmt's own exclusion pattern so KSP
output under build/ is left alone.

Cleanup

Categories, Ingredients and Areas had become three copies of one design.
rememberListUi holds the StoreReadResponse → Loading/Content/Failure mapping
once — deliberately not a CircuitUiState, since Circuit pairs a state with
exactly one UI, so each tab keeps its own state type and only the translation is
shared. produceStoreState does the same for the producers. Net −320/+128, and
the Error.Custom TODO exists in one place instead of three.

Known limitation

list.php?a=list returns ~195 areas, but filter.php?a= only matches the exact
strArea string meals actually carry — 14 distinct values on v1. Most rows in
the Areas tab therefore have no recipes behind them, and asking for both
spellings recovers some but not all of that gap. Recorded in the themealdb-api
skill so an empty area is not mistaken for a bug in the screen. Deciding what the
tab should show is left for a follow-up.

Verification

ktfmtCheck + detektAll clean, allTests green (85 tests, 11 new covering
both new presenters), all six targets for every library module, plus
:app:assembleDebug, :desktopApp:build and :webApp:build so every Metro graph
is exercised. Each of the nine commits was compile-checked individually. Manually
walked the desktop app from a wiped database: all three tabs populate, British
returns 15 recipes, back restores from cache, no UI errors.

🤖 Generated with Claude Code

https://claude.ai/code/session_018NME4RvM9ESMNfA3HReTKG