An existing Vite+ project's version-update prompted to set up pre-commit hooks
whenever a legacy husky/lint-staged config was present, unlike agent/editor
setup which already defer to --full. For an unsupported lint-staged format the
prompt could never do anything (it always ended in "skipping git hooks setup").

getExistingVitePlusSetupOptions now defaults hooks to false like agent/editor;
legacy husky/lint-staged migration runs only with `vp migrate --full` or an
explicit `--hooks`. Drops the now-unused legacyGitHooksMigrationCandidate param
and variable.

Claude-Session: https://claude.ai/code/session_01DQhS6o1fyQd1yjiee6W8jR

…igrate.sh

The version report resolved the package manager via `vp env current | vp node`,
but the isolated install runs with VP_NODE_MANAGER=no, so `vp node` has no
managed Node and the parse returned empty: no `-r`, so `vp why` queried the
workspace root without recursion and reported nothing.

Detect pnpm from the project's own markers (a `packageManager` pin, a workspace
file, or the post-install lockfile) instead. For pnpm, report all three packages
in one recursive query (`vp why -r vite-plus vite vitest`); npm/yarn/bun still
loop one package at a time since their `why` accepts only one.

Claude-Session: https://claude.ai/code/session_01DQhS6o1fyQd1yjiee6W8jR

Two fixes to the vitest-injection guard in rewritePackageJson (PR review):

- The injection fallback matched any dep whose name contains "vitest",
  including VITEST_DIRECT_USAGE_EXCLUDED packages like @vitest/eslint-plugin.
  When such a package drove the only signal, a catalog monorepo omitted the
  vitest catalog entry yet this still injected `vitest: catalog:`, a dangling
  spec that breaks pnpm install. Use isVitestAdjacent (which applies the
  exclusion filter).
- A @nuxt/test-utils package keeps its `from 'vitest'` imports (the rewriter
  preserves them for Nuxt), but the guard never added a direct vitest, so under
  strict pnpm/Yarn layouts the preserved imports could not resolve. Add
  hasNuxtTestUtilsDependency to needDirectVitest and the injection condition.

Adds two reproducing migrator tests.

Claude-Session: https://claude.ai/code/session_01DQhS6o1fyQd1yjiee6W8jR

Correctness fixes from the xhigh workflow review of this branch vs main:

- vite-plus-bootstrap: write the pnpm-workspace.yaml catalog (creating the
  file when missing) on standalone pnpm 9.5-10.6.1 upgrades so reconciled
  catalog: specs resolve, and judge catalog: specs/overrides through the
  workspace catalog in the <10.6.2 pending check so a migrated project
  converges to the already-using-Vite+ fast exit
- compat runner: kill the config-resolution worker after 30s (new
  runCommandSilently timeoutMs) instead of hanging vp migrate when a
  project-local vite-plus predates the lazyPlugins-skip handshake
- npm reinstall: move stale node_modules/vite aside and restore it when the
  final npm install fails, instead of deleting it up front
- format: only fall back to whole-project formatting on a genuine "not a
  git repository"; dubious-ownership/git-missing now skip formatting; cap
  format arg batches at 24KB on Windows (32,767-char CreateProcess limit)
- catalog: survive scalar-valued peerDependencyRules entries in
  pnpm-workspace.yaml instead of crashing mid-write
- import rewriter: rewrite declare module 'vite' augmentations in every
  file again (they must track the rewritten config import identity) while
  keeping the #2004 config-only scoping for value imports and reference
  directives; count triple-slash-only preserved vitest references in
  preserved_vitest_files
- package detection: verify an ancestor .pnp.cjs actually owns the project
  (findPackageLocator) before running setup(), so foreign PnP hooks are
  never installed process-wide
- preview registry: stash a custom Yarn Berry npmRegistryServer in a YAML
  comment and restore it on real-release cleanup instead of deleting it
- prompts: stop interpolating huge path batches into the vp fmt failure hint
- oxlint plugin: bypass the @nuxt/test-utils cache when stat fails instead
  of colliding on the mtimeMs 0 sentinel

…import

Regenerate the migration-rewrite-declare-module snapshot for the review fix
that rewrites declare module 'vite' augmentations in every file again: an
augmentation binds to the literal specifier, so it must target the same
module as the rewritten vite.config.ts import ('vite-plus') or the augmented
UserConfig options fail type-checking. Value imports and reference-type
directives keep the #2004 config-files-only scoping, and vitest augmentations
are still never rewritten.

Also document the augmentation exception in migrate-rules.md.

Partially reverts e29973c. Through the core alias a preserved
declare module 'vite' augmentation reaches the same
@voidzero-dev/vite-plus-core module whose UserConfig types defineConfig
from vite-plus, so it keeps working after migration. vite-plus exports
no UserConfig symbol, so the rewritten declare module 'vite-plus'
augmentation merged with nothing and silently stopped applying. Users
extending vite-plus's own surface write declare module 'vite-plus'
directly.

Restores the #2004 config-files-only scoping for all vite specifiers,
flips the regression test to assert preservation, regenerates the
migration-rewrite-declare-module snapshot, and documents the rationale
in migrate-rules.md.

…dpoint)