Accept a bare SharedArrayBuffer chunk in TS TextDecoderStream

[jasnell]

The Encoding Standard's TextDecoderStream writable accepts an
AllowSharedBufferSource, and Node decodes a bare SharedArrayBuffer
chunk. TS rejected it ("TextDecoderStream: chunk must be a
BufferSource") and errored the stream, though it already decoded views
over shared memory, and TextDecoder.decode() accepts a bare one. C++
rejects it too, as not a BufferSource.

The TS transform now accepts a SharedArrayBuffer, growable ones
included, and decodes it like an ArrayBuffer; a character split across
two of them decodes whole. C++ is unchanged, so this is a new
implementation divergence (encoding ledger #8).

Pins: encoding chunk-types.js decoderSharedArrayBufferChunks (parity
for shared-memory Uint8Array, subview and DataView chunks with a split
character; ledger #8 for bare, growable and split bare chunks: TS
decodes 'CD€', C++ rejects with its invalid-chunk TypeError). Without
the source change, it fails, in the TS cells only.

[ask-bonk]

Since last review: 0 resolved, 0 still open, 0 new.
LGTM!

Not re-run: tests, api-compat, docs, compat-flags, design-simplicity (no author changes in their files since the last review)

---

_{Reviewed commit: bb41075d · github run}

[guybedford]

Verified the change end to end: the Encoding Standard's writable chunk type is AllowSharedBufferSource, and the downstream TextDecoder::decode(kj::Array<const byte>) already unwraps a SharedArrayBuffer via jsg's tryUnwrap (value.h), so the cast-through is sound and growable SABs take the same path. All five encoding variants pass locally, and reverting only encoding.ts makes encoding-ts@ fail on decoderSharedArrayBufferChunks, so the pin is a real regression guard.

Two non-blocking nits:

1. In the decode() helper in chunk-types.js, in the C++ branch the reader.read() loop throws before await writes is reached, so the Promise.all in writes is left as an unhandled rejection. The harness swallows it, but a writes.catch(() => {}) (or try { ... } finally { await writes.catch(() => {}) }) would keep it clean and make explicit that the read-side error is the one asserted.
2. The TS message "chunk must be a BufferSource" and the AGENTS.md "Accepts BufferSource chunks only (...)" are now slightly imprecise, since a SAB isn't a BufferSource. The message is pinned under ledger #4 so I wouldn't change it here; AllowSharedBufferSource would be the exact term in the doc.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 38.59%. Comparing base (ae4c08a) to head (bb41075).

Additional details and impacted files

@@                           Coverage Diff                            @@
##           jasnell/ts-streams-identity-write-edges    #7643   +/-   ##
========================================================================
  Coverage                                    38.59%   38.59%           
========================================================================
  Files                                          868      868           
  Lines                                       268548   268548           
  Branches                                     25414    25414           
========================================================================
  Hits                                        103635   103635           
  Misses                                      150701   150701           
  Partials                                     14212    14212           

☔ View full report in Codecov by Harness.
📢 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.