feat: replace peek-stream with native node:stream

[hardik-kaji]

Why

The peek-stream package depends on through2 and xtend — all of which are
legacy userland stream utilities. In particular, xtend is no longer maintained.
Removing peek-stream reduces the dependency tree, avoids maintenance/security risks
from unmaintained packages, and aligns with the ongoing effort (#386) to replace
userland stream helpers with native node:stream APIs.

Summary

• Removes the peek-stream dependency by replacing it with a native createPeekStream
  implementation using node:stream Duplex
• Fixes a bug where onpeek received the entire buffered data instead of only the
  first maxBuffer bytes when a single write exceeded maxBuffer
• Moves createPeekStream to lib/utils.js with JSDoc documentation and comprehensive
  unit tests

Changes

• index.js — Remove inline createPeekStream, import it from lib/utils.js, remove
  unused Duplex import
• lib/utils.js — Add createPeekStream with proper peek slicing logic and JSDoc
• test/utils.test.js — Add 6 unit tests covering peek correctness, chunked input,
  short streams, error propagation, empty streams, and transform destination

Checklist

• run npm run test && npm run benchmark --if-present
• tests and/or benchmarks are included
• documentation is changed or added
• commit message and code follows the Developer's Certification of Origin
  and the Code of conduct

[fritzfelix]

Thank you for this. Unfortunately those EOL packages lead to quite some headaches within corporate environments, thus it's much easier to use fastify in those environments if it's clean.

[mcollina]

This seem expensive perf-wise. Maybe we can drop peek-stream and have a better interface for this?

[hardik-kaji]

This seem expensive perf-wise. Maybe we can drop peek-stream and have a better interface for this?

@mcollina Done! Refactored createPeekStream into createLazyTransform with a simpler, more performant interface:

• Transform instead of Duplex — uses idiomatic transform/flush instead of manual write/final/read with
  event listener wiring
• Synchronous choose(data) instead of callback-based swap(err, stream) — the caller just returns the
  desired stream directly, no callback overhead
• Array-based buffering — chunks are collected in an array and Buffer.concat is called only once when the
  threshold is reached (instead of on every chunk)
• Default maxBuffer = 10 — since both callers always pass 10, it's now the default

The callers are now much cleaner:

  function zipStream (deflate, encoding) {
    return createLazyTransform(function (data) {
      switch (isCompressed(data)) {
        case 1: return new Minipass()
        case 2: return new Minipass()
      }
      return deflate[encoding]()
    })
  }