fix(update): warn when gateway is running before hermes update

[ygd58]

Problem

Running hermes update while the gateway is active can leave Telegram polling in a broken state. Old process holds the bot token, new binary cannot re-bind it.

Fix

Add a pre-update gateway check: if running PIDs are found, warn the user and prompt to confirm. Recommended flow: hermes gateway stop → hermes update → hermes gateway start.

Fixes #23783 (partial)

[teknium1]

This looks implemented on current main via the newer update/restart and Telegram conflict-recovery path.

Automated hermes-sweeper review evidence:

• hermes_cli/main.py:8898 now has a post-update gateway restart block for all running gateways, because the code update is shared across profiles.
• hermes_cli/main.py:9299 handles manual non-service gateways: it finds running gateway PIDs, relaunches profile gateways where possible, signals old processes, waits briefly for exit, and prints manual restart guidance for unmapped processes.
• hermes_cli/gateway.py:557 documents find_gateway_pids(..., all_profiles=True) specifically for hermes update, since updating code affects every profile.
• gateway/platforms/telegram.py:1074 recovers the post-update Telegram polling-conflict case where the old long-poll session is still held server-side.
• gateway/platforms/telegram.py:1703 clears stale webhook state before polling and routes polling conflicts into that recovery path.
• The Telegram conflict recovery landed in f260aa6dc0fd6ecafb36964c33a39276fdd8da0a and is contained in v2026.5.28 and later tags.

The exact UX from this PR — warning before update and asking the user to manually stop/start — is not the path main chose, but the underlying bug this PR was trying to avoid is covered by the current automatic restart/conflict-recovery behavior.