ultracite@7.9.3

Patch Changes

• dc6d760: Disable the n/no-unpublished-import rule in the ESLint core config. This rule flags imports of packages that aren't listed as published dependencies, but it produces a lot of false positives in practice, so it's now turned off.
• 15ecfba: Fix ultracite init --linter eslint installing an unusable toolchain. The generated ESLint config imports eslint-plugin-storybook unconditionally (which requires the storybook peer) and the generated Stylelint config extends stylelint-config-standard / stylelint-config-idiomatic-order / stylelint-prettier, but none of those packages were installed — so a fresh ESLint setup failed to load with "Cannot find package 'storybook'" or "Could not find stylelint-config-standard". These four packages are now installed with the ESLint linter.
• 2f73a41: Upgrade to oxlint 1.73.0 and oxfmt 0.58.0, and enable the new lint rules they introduce: no-unreachable-loop, unicorn/explicit-timer-delay, and unicorn/no-confusing-array-with.
• 83b2783: Add an [astro] formatter mapping (astro-build.astro-vscode) to the oxlint VS Code editor settings generated by ultracite init, since oxfmt doesn't format .astro files.
• d186c53: Move every JS-plugin-based rule set out of the Oxlint core and framework presets and into a single opt-in ultracite/oxlint/js-plugins preset. This covers eslint-plugin-github and eslint-plugin-sonarjs (previously in core) as well as oxlint-plugin-react-doctor (previously bundled into the react, next, and tanstack presets). The core, react, next, and tanstack presets now run entirely on Oxlint's native Rust rules, so new setups no longer install those dependencies and no longer pay the slower JS-plugin lint pass. To keep the extra ESLint-parity and React Doctor rules, install eslint-plugin-github, eslint-plugin-sonarjs, and oxlint-plugin-react-doctor and extend ultracite/oxlint/js-plugins alongside core (and your framework preset).
• 057753e: Add performance benchmarks for ultracite check / ultracite fix across all three providers (oxlint, biome, eslint). A new CI job builds the PR and its base branch on the same runner, benchmarks them interleaved, and fails on a statistically significant regression (median ratio > 1.25x with Mann-Whitney U p < 0.05) so config changes can't silently slow the linters down again.