refactor(core): reuse Markdown chunk byte counts

[kitlangton]

Why

HTML-to-Markdown rendering encodes each appended chunk to measure its UTF-8 size, then encodes the same string again when it fits unchanged. The second encoding allocates a redundant byte array for the common path.

What Changes

Reuse the first byte count when the chunk fits. If truncation changes the string, continue measuring the retained prefix as before. Output text, byte budgets, and room reserved for closing syntax are unchanged.

Scope

One production expression and three boundary cases in the existing webfetch tests. No tool contracts or parser chunking changes; this does not address the existing surrogate-pair splitting behavior at parser chunk boundaries.

Verification

cd packages/core
bun run test test/tool-webfetch.test.ts
bun typecheck
cd ../..
bunx prettier --check packages/core/src/tool/html-markdown.ts packages/core/test/tool-webfetch.test.ts
git diff --check

57 tests passed with 102 assertions. Typechecking, formatting, and whitespace checks passed. Added cases cover fitting and truncated ASCII, two-byte, and four-byte UTF-8 content, with surrogate pairs aligned to parser chunks. No throughput benchmark claim.