fix: give client GameTests a virtual clock/dispatcher instead of real threads

[KP2048]

Summary

Follow-up to #9's "known open issue" note: gives client GameTests a virtual clock/dispatcher instead of real threads/wall-clock time, fixing the ~1-in-5 intermittent failure rate documented there.

Root cause

ComposeScreen/ComposeContainerScreen back their coroutine scope with Dispatchers.Default and real wall-clock time. A composable's delay(...) (e.g. a dialog's close animation) genuinely raced real thread scheduling and frame delivery against the harness's tick-based polling - a different test failing each run, no logged exception.

Fix

AClientGameTestHarness.run() now installs a ComposeTestClockOverride (common/.../gui/ComposeScreen.kt) around each @ClientGameTest: a StandardTestDispatcher plus a per-frame scheduler.advanceTimeBy(50) pump called from renderNodes(), matching how real Jetpack Compose's own test tooling (ComposeTestRule/runComposeUiTest) avoids this class of race by construction.

Real gotcha hit and fixed along the way: the first attempt used advanceUntilIdle() instead of advanceTimeBy(bounded). Composables can run legitimately infinite delay() loops (TextFieldCore's blinking-cursor LaunchedEffect), and advanceUntilIdle() only returns once truly nothing is scheduled anywhere - never, for an unboundedly-recurring loop. That hung the render thread permanently the moment any text field was on screen, and every subsequent test in that run then failed too (client's main-thread executor queue never got to run again). Switched to a fixed, bounded per-pump increment instead.

Keeping the dependency out of the shipped jar

kotlinx-coroutines-test is dev/test-only: compileOnly in common/build.gradle.kts, runtimeLibrary(...) (present for local runs like runGametestClient, never bundled - and the config that correctly handles NeoForge's classloader-layer artifact transform, not plain runtimeOnly) in the loader modules. ComposeScreen itself never references kotlinx.coroutines.test.* symbols directly - it only holds a plain CoroutineDispatcher? and a (() -> Unit)? pump callback (already-bundled core/stdlib types) - so a real player's game, which never has the test dependency on its classpath, never needs to resolve it.

Also in this PR

• Merges origin/1.21.x (now including fix: NBT serialization crash in ArchieFluidSlot/Storage and ArchieEnergyStorage #9) - resolved one AGENTS.md conflict by replacing fix: NBT serialization crash in ArchieFluidSlot/Storage and ArchieEnergyStorage #9's "known open issue, cause not confirmed" note with this PR's "confirmed fixed" one.
• Fixes two experimental-API opt-in warnings surfaced along the way: ArchieFluidSlot.Serializer's encodeNullableSerializableElement/decodeNullableSerializableElement (from fix: NBT serialization crash in ArchieFluidSlot/Storage and ArchieEnergyStorage #9) need @OptIn(ExperimentalSerializationApi::class), matching the existing pattern elsewhere (NetworkChannel.kt, NBT.kt); AClientGameTestHarness.run()'s new advanceTimeBy(...) call needs @OptIn(ExperimentalCoroutinesApi::class).

Test plan

• :common:compileKotlin, :fabric:compileKotlin, :neoforge:compileKotlin all clean, no warnings
• kotlinx-coroutines-test confirmed resolving correctly via runtimeLibraryConfiguration
• Full local neoforge:runGametestClient: 8 consecutive clean runs, 131 individual tests total, zero failures - including the previously-hanging text field test and the originally-flaky ConfirmDialog/RadioGroup tests
• Caught and fixed a real regression before landing: the first advanceUntilIdle() attempt hung the entire suite on the text field test; verified the advanceTimeBy fix resolves it before re-running the full batch

🤖 Generated with Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.