fix(readable): install a StringDecoder when setEncoding is called

[SAY-5]

What kind of change does this PR introduce?

bug fix

What is the current behavior?

BodyReadable.setEncoding only wrote _readableState.encoding and skipped the Node-standard step of installing a StringDecoder. With an encoding set but no decoder, the Readable machinery falls through to Buffer.prototype.toString(encoding) on each individual chunk, which corrupts any multi-byte character (UTF-8 CJK, emoji, any 3+ byte sequence) that straddles a chunk boundary. The partial bytes become U+FFFD, and the next chunk's leading bytes do too.

Fixes #5002.

What is the new behavior?

Match Node's Readable.prototype.setEncoding: build a StringDecoder from the supplied encoding and wire it into _readableState. The decoder buffers unfinished byte sequences between chunks, so callers iterating for await (const chunk of body) with setEncoding('utf8') get lossless text.

Other information

Added test/readable-set-encoding-utf8.js: pushes the 3-byte sequence for 中 split across two chunks and asserts no U+FFFD in the decoded output. Pre-fix the test fails with "\uFFFD\uFFFD"; post-fix the decoded string is exactly "中".

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.13%. Comparing base (2a6f9c7) to head (6db976f).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #5086   +/-   ##
=======================================
  Coverage   93.13%   93.13%           
=======================================
  Files         110      110           
  Lines       35816    35824    +8     
=======================================
+ Hits        33356    33364    +8     
  Misses       2460     2460           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[mcollina]

Can you check why was this done in this way? Also, why are we overriding setEncoding now? Shouldn't removing the override do the same?

[SAY-5]

Good point @mcollina — you're right that removing the override is the cleaner fix. node:stream's Readable.prototype.setEncoding already installs a StringDecoder and buffers partial byte sequences between chunks; the pre-existing override only set _readableState.encoding and nothing else, which is exactly what caused #5002. Force-pushed to just delete the override (and the now-unused node:string_decoder import). Regression test still passes.