Fix dev supervisor readiness after UTF-8 log output

[hanafusay]

Summary

• treat readiness log offsets as byte offsets
• preserve readiness detection after earlier UTF-8 output
• cover the behavior through the real child process lifecycle

Problem

The supervisor records a log position using the file size in bytes, but readiness polling applied that value as a JavaScript string character offset. Earlier multibyte UTF-8 output could therefore cause a newly appended readiness marker to be skipped.

Verification

• node --test test/dev-supervisor-child.test.ts
• npm run typecheck
• targeted ESLint and Prettier checks

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}