[Bug]: list reporter corrupts test stdout/stderr that is not valid UTF-8 per chunk (multi-byte chars split across writes) #43079

Version

main @ 48688e0 (1.64.0-next); the code paths are unchanged in 1.63

Steps to reproduce

import { test } from '@playwright/test';

test('split multi-byte char', async () => {
  // "€\n" = e2 82 ac 0a, written in two chunks
  process.stdout.write(Buffer.from([0xe2, 0x82]));
  process.stdout.write(Buffer.from([0xac, 0x0a]));
});

Run with npx playwright test --reporter=list.

Expected behavior

The terminal shows €, as it did before #42886, because the reporter wrote the raw bytes through to the stream.

Actual behavior

The terminal shows �� (two replacement characters). This happens with or without printWorkerIndex.

Since #42886 (printWorkerIndex), ListReporter._dumpToStdio does chunk.toString('utf-8') and writes the decoded string instead of chunk. Worker output reaches the reporter as Buffers (stdioChunkToParams sends Uint8Array chunks as base64), so each chunk is decoded on its own. A character split across two writes is lost, and so is any output that isn't UTF-8, such as Latin-1 from a child process. Decoding the two chunks separately gives "��\n", while the concatenated bytes give "€\n".

Additional context

Suggested fix: when there is no worker prefix to add, write the original chunk unchanged. When prefixing, decode with a per-stream StringDecoder (node:string_decoder) so partial sequences carry over to the next chunk. Low severity; found during a code review.

Environment

Not environment-specific: the code is platform-independent.
Verified by running the functions from packages/playwright-core/src/server (or the reporter logic) directly with Node 22.22 on Linux x64.
No browser is needed to reproduce.