[kevin9327]

What

redactTelemetryString no longer leaves half of a surrogate pair at the cut.

truncateTelemetryString cut at maxLength - 1 UTF-16 code units. When that position fell between the two halves of an astral character (emoji, many CJK extension characters), the kept text ended in a lone high surrogate before the …. That string is not well-formed UTF-16: encodeURIComponent throws URIError on it, and TextEncoder/JSON serialisation turn it into U+FFFD, so a telemetry or observation message with an emoji near the 240-character limit either breaks the code that sends it or reaches the backend with a replacement character.

redactTelemetryString(`render failed ${"🎬".repeat(20)}`, 18)
// before: "render failed 🎬\uD83C…"   (lone high surrogate)
// after:  "render failed 🎬…"

How

If the last kept code unit is a high surrogate, the cut moves back by one, so the pair is dropped whole. Strings under the limit and cuts that land on a character boundary are unchanged; the result is never longer than maxLength.

Test

New case in packages/core/src/telemetryRedaction.test.ts: a message of emoji truncated at a limit that falls inside a pair keeps only whole characters and is isWellFormed(). The whole file passes (50 tests). Formatted with oxfmt 0.41.0.

fails without the fix: packages/core/src/telemetryRedaction.test.ts › "does not cut an astral character in half when truncating" (ends in \uD83C…)

This PR was prepared with AI assistance; I reviewed the change and ran the tests.