Feature request: de-emphasized/subtle text support for assistant-generated markdown

Problem

Currently, there's no supported way for the assistant's response text to render as visually de-emphasized (e.g., dimmed, muted) — the existing subtle theme token only applies to built-in UI elements (borders, secondary chrome), not to text the model itself generates.

Use case

Appending metadata (e.g., timestamps) to responses that should be present and scannable, but easy to visually skip past unless specifically looked for — without being visually prominent by default. This generalizes beyond timestamps to any secondary/skippable content a model might want to include (footnotes, disclaimers, etc.).

Proposed solution

A markdown extension (e.g., a custom span/inline syntax) that lets assistant-generated text opt into the existing subtle theme token, rather than requiring literal ANSI color codes. This would automatically respect whatever a user's active theme defines as "subtle," working correctly across light/dark themes and custom themes without the model needing to guess at colors that might not render consistently or might get stripped as unsupported input.

Why this over raw ANSI passthrough

Theme-token-based de-emphasis stays consistent regardless of the user's theme; raw ANSI color codes chosen by the model would not.