fix: remove auto-install feature from local CLI for better performance

[fengmk2]

The vp run command was automatically running pnpm install before
executing tasks, causing serious performance degradation. Remove the
entire auto-install feature:

• Delete create_install_synthetic_request() and auto-install block
  from execute_vite_task_command() in the NAPI binding
• Remove VITE_DISABLE_AUTO_INSTALL env var from 147 snap test files
• Delete auto-install-before-task snap test and benchmark script
• Update docs and RFCs to remove auto-install references

[fengmk2]

• fix: remove auto-install feature from local CLI for better performance #631 👈 (View in Graphite)
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[fengmk2]

Here are the benchmark results using vp run lint on main (hyperfine, 20 runs):                       
                                                                                                       
  Cold cache (after vp cache clean)                                                                    
                                               
  ┌─────────────────────────────────────────────────────┬─────────┬─────────┬─────────┐                
  │                      Scenario                       │  Mean   │   Min   │   Max   │
  ├─────────────────────────────────────────────────────┼─────────┼─────────┼─────────┤
  │ Auto-install enabled (main default)                 │ 2,168ms │ 1,817ms │ 2,633ms │
  ├─────────────────────────────────────────────────────┼─────────┼─────────┼─────────┤
  │ Auto-install disabled (VITE_DISABLE_AUTO_INSTALL=1) │ 610ms   │ 462ms   │ 1,241ms │
  └─────────────────────────────────────────────────────┴─────────┴─────────┴─────────┘

  Auto-install overhead: ~1.56s per invocation (3.6x slower)

  Warm cache (task cache hot)

  ┌───────────────────────┬───────┬───────┬─────────┐
  │       Scenario        │ Mean  │  Min  │   Max   │
  ├───────────────────────┼───────┼───────┼─────────┤
  │ Auto-install enabled  │ 524ms │ 414ms │ 1,201ms │
  ├───────────────────────┼───────┼───────┼─────────┤
  │ Auto-install disabled │ 448ms │ 383ms │ 1,029ms │
  └───────────────────────┴───────┴───────┴─────────┘

  Auto-install overhead: ~77ms per invocation (1.17x slower) — the install result itself gets cached by
   the task runner, so warm overhead is lower.

  Raw baseline

  ┌───────────────────────────────────┬──────────────┐
  │            Measurement            │     Time     │
  ├───────────────────────────────────┼──────────────┤
  │ pnpm install (already up to date) │ 1,650ms mean │
  └───────────────────────────────────┴──────────────┘

  The cold cache overhead matches the raw pnpm install cost almost exactly. On warm cache, the task
  runner caches the install result, but there's still ~77ms overhead for cache lookup + package manager
   resolution. Note: the original benchmark script (bench/benches/auto_install.sh) was testing vp lint
  which goes through execute_direct_subcommand and was never affected by auto-install — so it would
  have shown no difference.

[fengmk2]

Merge activity

• Feb 27, 2:32 AM UTC: A user started a stack merge that includes this pull request via Graphite.
• Feb 27, 2:33 AM UTC: @fengmk2 merged this pull request with Graphite.