Charset-aware, binary-safe body previews in request/response logging

[OmarAlJarrah]

Problem

Body previews hardcode UTF-8 (utf8Preview, DefaultInstrumentationStep.kt:323) even though MediaType.charset is parsed and cached (MediaType.kt:53). A latin-1 body logs as mojibake; a gzip/binary body logs as U+FFFD noise.

Proposed change

Add previewText(bytes, mediaType) honouring the declared charset, plus a small isProbablyText heuristic (sample the first N code points) to skip previews for binary bodies. Land it in the shared emitter (see the emitter-extraction issue).

Acceptance

• Declared charset respected
• Binary bodies not previewed as text
• Tests for latin-1 + gzip

Dependencies

• Blocked by Extract a shared instrumentation emitter from the sync and async steps #26 — land the charset/binary-sniff logic in the shared instrumentation emitter

---

Priority: high · Effort: small