text/html responses corrupted (mojibake) — both Edge Functions and Storage public objects

Text/html responses served by this project are corrupted in a way consistent with UTF-8 bytes being reinterpreted as Shift_JIS: raw HTML is rendered as literal text with garbled multi-byte (Japanese) characters instead of being parsed by the browser. This reproduces via two independent, unrelated serving paths on the same project, which points to a project/CDN-level charset or Content-Type handling issue rather than anything in our own code: Edge Functions — a minimal from-scratch Edge Function returning new Response(html, {headers: {'content-type': 'text/html; charset=utf-8'}}) exhibits the corruption. Storage — a public Storage bucket object (.html file) served via its public object URL exhibits the same corruption. The underlying content is confirmed correct UTF-8 in the database (verified via direct SQL query), so the corruption happens in how the response is served, not in how the data is stored. Project ref: eqlqturbagovcnngpipz Steps to reproduce: Deploy any Edge Function returning text/html with UTF-8 multi-byte characters (e.g. Japanese) and a content-type: text/html; charset=utf-8 header. Fetch the function's URL in a browser — the raw source is shown as text instead of being rendered, with multi-byte characters displayed as mojibake. Separately, upload the same HTML file to a public Storage bucket and fetch its public object URL — same result. A support ticket referencing this same project has also been filed via the dashboard, in case that's a faster path. Happy to provide more repro details (exact function code, screenshots) if useful.