feat: Obsidian VIP premium theme, English README, and PR APK builds

[ChimeraAnt-DEV]

Summary

Three things in one PR: a new premium "VIP" theme, an English README aligned with the current upstream, and CI that actually builds an APK on pull requests.

---

1. Obsidian VIP premium theme (new)

A new OBSIDIAN_VIP colour theme — deep obsidian surfaces, champagne-gold primary, violet and aurora-cyan accents — designed to read as "premium" while staying reactive to theme changes.

Area	Change
ui/theme/PremiumTheme.kt	New. obsidianVipDark scheme, LocalPremiumTheme, premiumGoldGradient(), rememberPremiumSheenBrush(), premiumBorderStroke(), Modifier.premiumAmbience(), PremiumBadge()
ui/theme/Theme.kt	Resolves OBSIDIAN_VIP, forces dark in both branches, provides LocalPremiumTheme
ui/theme/ColorTheme.kt, NativeThemeUtils.kt	Enum entry + seed colours
ui/screens/main/MainScreen.kt	VIP badge beside the title, ambience overlay layer
ui/components/BackgroundCard.kt	Picks up the gold gradient border automatically
LauncherSettingsScreen.kt, all 3 strings.xml	Theme entry and labels (en / zh-CN / zh-TW)

The premium effects are driven by Compose animation primitives and sit behind LocalPremiumTheme, so they cost nothing when another theme is active. The card border is a static gold gradient and the only animated element is the single ambience layer, to keep GPU cost predictable.

2. README → English, aligned with upstream

README.md is now English and mirrors the current upstream structure (About, Language/Translation, Build, License, Open Source Libraries) with the library table refreshed. The Chinese original is preserved as README_ZH_CN.md; README_EN_US.md is removed as redundant. Language cross-links updated in all three files.

3. CI: APK builds on pull requests

build.yml previously had no pull_request trigger and the job was explicitly gated with if: github.event_name != 'pull_request', so a PR would show no build at all. Now:

• pull_request trigger added; the gating condition is removed.
• On a PR, one architecture (all) is built — enough to prove the code compiles. Manual and release runs still build all five.
• Added a concurrency group so superseded pushes to the same PR cancel their predecessor runs.
• Added permissions: contents: read and a VARIANT env with a Debug default, since inputs.variant is empty on pull_request.
• push_ci.yml narrowed from branches: '**' to main, so the same commit isn't built twice once PR builds are live.

4. GPLv3 fork hygiene

gradle.properties now names the fork Java Ant Launcher, points url_home at this repository, and the splash screen shows an "unofficial modified build" notice (translated to zh-CN/zh-TW). The license's additional terms require a modified version to be distinguishable and marked as non-official; this is that.

---

What this PR deliberately does not do

The original request included removing the need to own Minecraft. I did not implement that, and want the reasoning on the record so it isn't silently expected:

• The ownership check is not a launcher setting. During Microsoft login the app calls Mojang's entitlements/mcstore endpoint, and online servers independently verify ownership server-side. Removing the client-side check would not enable multiplayer — it would only be a DRM bypass.
• Implementing it would be circumvention of a licensing/authentication control, which I'm not able to do.

The legitimate path for the stated use case (Bedrock players on Android who can't afford a second copy) already exists in the codebase: the launcher supports offline/local accounts (localLogin, OfflineYggdrasilServer) with no purchase and no login, which covers single-player worlds and mods. I documented this in a new Accounts section in the README with a comparison table and step-by-step instructions. If that documentation is wrong or incomplete, that's a normal review comment and I'll fix it.

---

Verification

The environment has no Android SDK/NDK, and the project uses a preview toolchain (AGP 9.3, Kotlin 2.4.10, compileSdk 37 with minorApiLevel 2), so I could not run a full Gradle build locally. This PR intentionally relies on CI to produce the first real APK — that is the point of the build.yml change.

What I did verify:

• Syntax-checked every modified Kotlin file with a standalone Kotlin compiler (JDK 21). No parse errors; the only diagnostics were unresolved references from the absent Compose classpath.
• Confirmed the exhaustive when (colorTheme) blocks in Theme.kt and the settings label mapping all handle OBSIDIAN_VIP.
• Validated all three strings.xml files as well-formed XML with no duplicate keys.
• Parsed all workflow YAML and checked the trigger/matrix/env structure resolves as intended.
• Confirmed no keystores or secrets were added to the diff (the two .jks files were already tracked upstream, untouched here).

Reviewer note: if CI fails, treat the build log as authoritative over my static checks — I had no way to compile Compose locally.

---

This PR was created by an AI agent (OpenHands) on behalf of the user.

@ChimeraAnt-DEV can click here to continue refining the PR

[ChimeraAnt-DEV]

CI verified. The pull_request trigger works and the APK builds cleanly:

• Run: #35175365966 — build (all) passed in 6m30s
• Artifact: debug apk (all) (394 MB universal debug, all ABIs, unminified — expected size for a debug build bundling LWJGL/SDL natives)

This is the first real compile of these changes, since my environment has no Android SDK/NDK. It confirms the OBSIDIAN_VIP enum wiring, the new PremiumTheme.kt file, and all three strings.xml edits compile and package correctly. Release artifacts and the per-architecture matrix are skipped on PRs by design; run the workflow manually (workflow_dispatch) to produce all five architectures.

---

This comment was created by an AI agent (OpenHands) on behalf of the user.