test(ui): cover app-shell dialogs, router guards, dialog gating, VM navigation/toggles

[tstapler]

Summary

Raises coverage of the ui/ app shell (backlog item f344219b): extracts dialog-selection predicates from GraphDialogLayer.kt and adds headless tests for dialogs, router guards, flashcards, and ViewModel navigation/UI toggles.

What Changed

• GraphDialogLayer.kt: internal fun AppState.shouldShow*() predicates (behavior-preserving, used by hosts)
• androidUnitTest: WikiSubdirFixDialogTest, FlashcardsScreenTest, ScreenRouterGuardTest, GraphDialogLayerHostsTest
• commonTest: GraphDialogLayerLogicTest
• businessTest: StelekitViewModelNavigationTest, StelekitViewModelUiToggleTest
• project_plans/ui-shell-test-coverage/ plan, research, validation (latest commit 56e3a3bd records the sdd:6-verify re-run)

Acceptance criteria

Passing (7): AC1 WikiSubdirFixDialogTest, AC2 FlashcardsScreenTest, AC3 ScreenRouterGuardTest, AC4 predicates + GraphDialogLayerLogicTest, AC6 UiToggle test, AC7 headless (no X11), AC9 deferred files documented.

Open (2) — not met as written:

• AC5 (history-size cap boundary): StelekitViewModel has no history cap; adding one is a product change. goBack/goForward boundaries are covered.
• AC8 (per-package floors ui/ ≥28% line, ≥8% branch, VM ≥48% branch): CI gate has no per-package thresholds; the existing diffCoverage --fail-under 80 gate was accepted.

Both are recorded as owner decisions in project_plans/ui-shell-test-coverage/validation.md lines 123–161. The backlog AC text was not reworded.

/backlog/review did not record a review this session: request_review rejects while AC5 and AC8 are not pass. A human needs to review and reconcile those two criteria.

Test plan

• :kmp:testDebugUnitTest — WikiSubdirFixDialogTest 6, FlashcardsScreenTest 4, ScreenRouterGuardTest 6, GraphDialogLayerHostsTest 6, GraphDialogLayerLogicTest 15: 0 failures
• :kmp:jvmTest — StelekitViewModelNavigationTest 7, StelekitViewModelUiToggleTest 6: 0 failures
• Remote CI: Bazel jobs and emulator smoke tests still pending/unreported at time of writing

[github-actions]

JVM Load Benchmark (Desktop)

Synthetic in-memory benchmark measuring load performance for the desktop (JVM) app.
Comparing fd7f0537 (this PR) vs 0db11151 (baseline)
Graph config: xlarge — 230 pages

Metric	This PR	Baseline	Delta
Phase 1 TTI ↓	0ms	0ms	0 (0%)
Phase 2 background ↓	0ms	0ms	0 (0%)
Phase 3 index ↓	0ms	0ms	0 (0%)
Total ↓	0ms	0ms	0 (0%)
Write p95 (baseline) ↓	30ms	27ms	+3ms (+11%) ⚠️
Write p95 (under load) ↓	1ms	n/a
Jank factor ↓	0.03x	n/a
↓ lower is better

Flamegraphs (this PR)

**Allocation** — object allocation pressure (JDBC/SQLite churn)

Alloc flamegraph not available

CPU — method-level hotspots by on-CPU time

CPU flamegraph not available

Top allocation hotspots (this PR)

`34.8%` byte[]_[k] `6.9%` java.lang.String_[k] `5.5%` int[]_[k] `5.4%` java.lang.Object[]_[k] `5.1%` java.util.LinkedHashMap$Entry_[k]

Top CPU hotspots (this PR)

`97.5%` /usr/lib/x86_64-linux-gnu/libc.so.6 `0.7%` /tmp/sqlite-3.51.3.0-7ddb2cb6-be06-4e50-b2c2-3b323a0e8d13-libsqlitejdbc.so `0.5%` fsync `0.3%` __libc_pwrite `0.1%` pthread_cond_signal

[github-actions]

Android Load Benchmark

Instrumented benchmark on an API 30 x86_64 emulator — 500-page synthetic graph.

Comparing fd7f0537 (this PR) vs 0db11151 (baseline)
Device: API 30 x86_64 emulator — 530 pages loaded

Graph Load

Metric	This PR	Baseline	Delta
Phase 1 TTI ↓	46ms	27ms	+19ms (+70%) ⚠️
Phase 3 index ↓	4611ms	2760ms	+1851ms (+67%) ⚠️

Interactive Write Latency (during Phase 3)

Metric	This PR	Baseline	Delta
Write p95 (baseline) ↓	8ms	8ms	0 (0%)
Write p95 (during phase 3) ↓	12ms	22ms	-10ms (-45%) ✅
Jank factor ↓	1.5x	2.75x	-1.25x (-45%) ✅
Concurrent writes ↑	22	14	+8ms (+57%) ✅

SAF I/O Overhead (ContentProvider vs direct File read)

Measures Binder IPC cost added by ContentResolver per readFile() call.
Real SAF via ExternalStorageProvider will be higher on device; this is a lower bound.

Metric	This PR	Baseline	Delta
Direct read / file ↓	0.0ms	0.0ms	0 (0%)
Provider read / file ↓	0.3ms	0.2ms	+0ms (+17%) ⚠️
IPC overhead ratio ↓	8x	7x	+1x (+14%) ⚠️
↓ lower is better · ↑ higher is better

[tstapler]

request_review refused this item three times with the same message (AC8 is index 7, 0-based, and the tool only accepts pass):

request_review rejected: 1 acceptance criteria are not yet marked pass via report_progress: 7: CI incremental coverage gate floor updated to ui/ line ≥ 28%, ui/ branch ≥ 8%, StelekitViewModel branch ≥ 48%; gate passes in CI (status=in_progress). Call report_progress for each remaining criterion, or report_blocked if genuinely stuck

Per the backlog fallback, shipping as a draft for human review. AC8 stays marked fail: CI gate is :kmp:diffCoverage with no per-package floors; a merged JVM+Android JaCoCo baseline is needed to set ui/ floors.

[tstapler]

Tracked by Stapler Squad backlog item: ssq://onyx.staplerhome.internal/backlog/v1/bl_01M4WBVTXRVSRYBF12131ES3D1
Host: host_01M0H2V3S43XCN5MGPZ4Z5Y3JX