Fix markdown rendering, add inline button menus, and fix duplicate cancel message on /followup redirect

[Copilot]

Three independent issues: markdown was never rendered in AI responses (missing parse_mode everywhere), no interactive menus existed for model selection or navigation, and /followup redirecting an active stream caused both the original handler and /followup to each send a cancellation message.

Markdown rendering

• Added _is_parse_error(exc: BadRequest) -> bool helper — single source of truth for detecting Telegram entity-parse failures, used consistently across all three fallback sites.
• _safe_edit_text and _send_chunks now accept parse_mode; on a parse error they retry with plain text so the message is always delivered.
• All AI-response send-sites now pass ParseMode.MARKDOWN: _stream_to_message final edit, _send() closures in company_command/build_command/autorun_command, and about_command, weather_command, company_roles_command, _do_build_followup, _weather_reminder_job.

Inline button menus

• /start: New welcome command with an 8-button inline keyboard covering all major commands (replaces about_command mapping).
• /settings (no args): Shows a model-selection inline keyboard instead of plain text — vision indicator (✅/📝), ★ on current selection, "Clear preference" button.
• model_select_callback handles set_model:<name> / set_model:clear presses.
• cmd_shortcut_callback handles cmd:<name> presses from /start — shows a usage hint or the settings keyboard inline.

Interrupt fix + streaming responsiveness

• Added _silenced_cancels: set[int]; /followup adds the user ID before calling task.cancel() so the original handler's CancelledError catch suppresses the duplicate "⛔ cancelled" message:

_silenced_cancels.add(user.id)
existing_task.cancel()
# ...
except asyncio.CancelledError:
    if user.id not in _silenced_cancels:
        await context.bot.send_message(..., text="⛔ Your in-progress request has been cancelled.")
    _silenced_cancels.discard(user.id)

• STREAM_EDIT_INTERVAL reduced 2.0 s → 1.0 s for faster live streaming updates.