[v6.x] fix(fetch): error the stream when decompression fails after onComplete

[mcollina]

This relates to...

Fixes #5999

Rationale

On v6.x (6.7.0 – 6.29.0, the line bundled with Node 22), a response that declares content-encoding: gzip but carries a body that is not valid gzip never settles text(), json() or a stream read when the body arrives in a later packet than the headers. The request's AbortSignal cannot end it either, so the promise hangs forever.

onComplete removes the fetch controller's terminated listener (onAborted, added in #2823) as soon as the raw bytes have arrived. Decompression happens later in the body pull loop, so when zlib fails (Z_DATA_ERROR) and the loop calls controller.terminate(err), nothing errors the response stream.

main fixed this in #3754 by no longer removing the listener in onComplete and registering it only once per fetch. That change was never backported.

Changes

• Backport of afeb626c (fix aborting Streams #3754) onto v6.x.
• Register the onAborted listener only once per fetch (guarded by fetchParams.controller.resume) and stop removing it in onComplete. Keeping the listener registered means a late decompression failure correctly errors the response stream; registering it once per fetch prevents the MaxListenersExceededWarning on many redirects (redirecting >11 times causes "MaxListenersExceededWarning" #1711) that the previous per-request registration caused.
• Add a regression test reproducing Node 22 fetch hangs forever on a broken gzip body #5999 to test/fetch/encoding.js. It times out without the fix.
• Include the "aborting a Stream throws" test from the cherry-picked commit in test/fetch/issue-1711.js.

Bug Fixes

• fetch: the body stream is errored when decompression fails after onComplete.

Verification

• The reproduction from Node 22 fetch hangs forever on a broken gzip body #5999 now rejects with terminated (Z_DATA_ERROR) instead of hanging.
• test/fetch/issue-1711.js (redirects + abort-stream) passes.
• Ran test/fetch/*.js with --expose-gc — all pass except pre-existing environment failures caused by missing dev deps (formdata-node, jsdom) and a Node version --max-http-header-size difference, which fail identically on the unpatched branch.
• standard lint passes on the changed files.

Status

• I have read and agreed to the Developer's Certificate of Origin
• Tested
• Benchmarked (optional)
• Documented
• Review ready
• In review
• Merge ready