webfetch always decodes as UTF-8, mojibake on non-UTF-8 sites

Describe the bug

webfetch decodes the response body with new TextDecoder(), which is always UTF-8. The charset parameter from the Content-Type header is read into contentType but never used for decoding, and HTML <meta charset> is ignored entirely.

Any page served as GBK/GB18030, Shift_JIS, ISO-8859-1, etc. comes back as U+FFFD replacement-character mojibake, so text/markdown extraction produces garbage for a large class of CJK and legacy sites.

Steps to reproduce

Fetch a URL whose server returns Content-Type: text/plain; charset=gbk with a GBK-encoded body. The tool returns �-filled text instead of the decoded characters.

Affected code

packages/core/src/tool/webfetch.ts — const content = new TextDecoder().decode(body).

Suggested fix

Extract charset from the Content-Type header, fall back to the HTML <meta charset> when the header omits it, then fall back to UTF-8 (and guard unsupported labels).

Environment

• opencode version: latest dev