Celeritas/Nothirium Extra

Changed

• Updated the development environment to Sumire-Labs/CleanroomDevEnv 6b24e5c and migrated the Gradle build and project-specific scripts to Kotlin DSL.
• Updated Gradle to 9.9.0 and GitHub Actions setup-java to 6.2.0.
• Enabled parallel builds and configuration caching, including production JAR verification.
• Moved Spotless configuration to gradle/spotless/ and removed Felt, following the template update. The project's LGPLv3 license and source notices remain in place.
• Updated release-note extraction to handle BOMs, CRLF line endings, and headings inside code fences. Stable-stage handling, MCP artifact filtering, and Windows CI coverage remain supported.
• Resolved the optional Iris shadow-pass API independently of renderer development dependencies. Camera-based optimizations stay disabled when the API cannot report the shadow-pass state.

Removed

• Removed Actinium integration, generated option adapters, its fast-item LOD hook, and the development dependency. Actinium supports Cleanroom 0.6.x; Celeritas Extra requires 0.7.x. Celeritas and Nothirium integrations remain supported.

Fixed

• Fixed background FPS limits below 20 falling short of the selected rate. For example, a 15 FPS limit could produce about 10 FPS. The scheduler now carries timing remainders forward and avoids catch-up bursts after stalls.
• Fixed a crash when the block-entity dispatcher receives a null tile. Profiling also respects the dispatcher's invalid-tile check.
• Applied TokenEnvoy replacements to all compiled Mod classes, including renamed entry points.
• Fixed false unresolved-token reports by checking class constant-pool text, and selected runtime JAR variants for embedded and shaded dependencies.
• Fixed Shift+Tab starting at the wrong field when no field is focused in the standalone list editor. It now selects the last field.
• Fixed VSync showing as enabled when the vanilla VSync setting is off and the saved adaptive-sync preference is on.
• Moved both item-frame LOD and vertex-cache invalidation to the render thread when models are rebaked, preventing concurrent cache access during asynchronous model baking.