feat(gateway): add clearer Telegram command approval decision prompts

[hellodoggo]

Summary

• Reframes Telegram command approval prompts as plain-English decision cards.
• Preserves the existing approval enforcement path; this does not change whether commands are allowed, denied, scoped, or checked.
• Keeps the raw command visible under optional technical details.
• Adds best-effort risk/context framing for common command shapes:
  • read-only web/API data fetched and parsed locally
  • GitHub repo fetch/clone for inspection
  • remote content piped into a shell/interpreter
  • destructive, config-editing, service-control, script, and multi-action shell commands
• Includes targeted tests for the prompt copy and existing inline approval-button flow.

Why

Manual approval prompts are easier to answer when they explain the action type and likely risk in user-facing language. Broad safety detectors can flag both low-risk data parsing and high-risk remote code execution; this prompt copy helps users distinguish those cases without weakening the actual approval gate.

Implementation notes

• The command-trait helper is UX-only and explicitly not a security boundary.
• Existing command approval enforcement remains in the approval/Tirith path.
• Telegram inline button behavior is unchanged.
• Raw command details remain visible for users who want to inspect the exact command.
• This starts with Telegram because the Telegram adapter already owns this inline approval prompt; the formatter can be shared later if other gateways need the same copy.

Test plan

python -m pytest tests/gateway/test_telegram_approval_buttons.py -q

Result locally:

23 passed in 1.19s