Package.json script tasks don't expose npm lifecycle env; cache-enabled tasks strip host-stamped values

从 vt_plan 中的 plan_spawn_execution 开始，检查 EnvFingerprints::resolve、DEFAULT_UNTRACKED_ENV 以及现有的 MARKER_ENV_NAME 插入。然后端到端阅读 vp_run_env fixture，包括其缓存用例。完成标准是：package.json 中的任务保留相关的 npm 生命周期环境，包括启用缓存的任务，并覆盖所报告的行为。

Summary

When vite-task spawns a package.json script, the child gets none of the npm lifecycle environment variables (npm_execpath, npm_config_user_agent, npm_lifecycle_event, npm_node_execpath, …) that npm, pnpm, and Yarn all set when they run scripts. Tools that detect their invoking package manager from that env — npm-run-all2/run-p being the common one — silently fall back to npm, which then fails devEngines.packageManager enforcement in pnpm projects.

Downstream report: voidzero-dev/vite-plus#2317 (vpr check → run-p → npm → EBADDEVENGINES). Reproduced there with a minimal project: pnpm run check sets npm_execpath=<pnpm>/bin/pnpm.mjs, npm_config_user_agent=pnpm/11.20.0 …, npm_lifecycle_event=probe for the child; the same script through the task runner gets none of them.

Why a vite-plus-side fix can't fully cover it

voidzero-dev/vite-plus#2385 stamps npm_execpath / npm_config_user_agent / npm_node_execpath / INIT_CWD into the process env before Session::init snapshots it. That works for plain scripts (which default to cache: scripts: false), but two pieces need the planner:

1. Cache-enabled tasks strip the stamped vars again. plan_spawn_execution runs EnvFingerprints::resolve(&mut spawn_envs, &cache_config.env_config) before spawning, and DEFAULT_UNTRACKED_ENV contains no lowercase npm_* name (matching is case-sensitive on Unix). So for cache: { scripts: true } or config-defined tasks (cached by default), the session-level stamp is filtered out and the npm fallback returns — only in the cached case, which no fixture currently exercises. The VP_RUN marker had this exact problem and is re-inserted after the filter, with a comment explaining why.
2. npm_lifecycle_event / npm_lifecycle_script are per-task values (script name / script body). Only the planner knows them; a session-level stamp can't provide them. pnpm sets both for every script it runs.

Suggested direction

• Re-insert the lifecycle names post-filter next to the existing MARKER_ENV_NAME insert in vt_plan (or add them to DEFAULT_UNTRACKED_ENV), so host-provided lifecycle env survives cache-enabled tasks.
• Optionally set npm_lifecycle_event (and npm_lifecycle_script) from the task itself when spawning package.json scripts, which vite-plus cannot do from outside.

Happy to send a PR for either or both if the direction sounds right — the vp_run_env e2e fixture looks like the natural place for coverage (its cached case already demonstrates the marker surviving the env filter).