Fix common issues - #512

- python 3.14 no longer creates a loop on demand
- older telethon releases then crashed with "no current event loop" at startup

- catch AuthKeyNotFound next to AuthKeyDuplicatedError instead of dumping a traceback
- re-raise for clients that don't exit, so vc_connection can drop a dead VC_SESSION

- python 3.12+ prints a SyntaxWarning for these at startup
- the strings stay the same; invalid escapes become errors in a later release

- akipy 1.x renamed async_akipy to async_akinator, so the games plugin failed to load
- require akipy>=1.2.4 so existing installs upgrade

- collect REDIS_URI and REDIS_PASSWORD too; out[3] and out[4] raised IndexError
- wait on the child processes instead of get_event_loop(), which fails on python 3.14
- exit with a message when no client is configured

- set an event loop before pyrogram is imported
- print the bot session with an f-string
- reinstall telethon with pip after uninstalling it in session.sh

- stickermanager caught StickerSetInvalidError, but the import is StickersetInvalidError
- cleanaction called eod without importing it
- beautify and weather logged through LOGS without importing it
- beautify matched timeouts via an unimported playwright module

- get_paste returns (ok, {link, raw, error}) now; the callback still used the old key
- tapping the paste button in plugin help always raised UnboundLocalError

- the "video as doc" assignment ran unconditionally after the streaming check
- notes, filters, snips and greetings skipped their upload path for every video

- json_parser falls back on raw bodies from third-party APIs (carbon, youtube scraping)
- eval ran whatever those servers returned

- get() wrapped the None from an unset key into a list, so empty lists looked non-empty
- fresh installs got phantom entries in pmpermit, broadcast, nightmode and other lists
- add() then saved [None, item] back to the database

- apscheduler 3.11 looks up the running loop in start(), and plugins start it at import
- nightmode jobs never ran, and autopic failed to load after a restart

- the README runs sessiongen after installing requirements, inside the bot's venv
- the reinstall pulled the newest telethon back in, past the <1.45 cap
- ssgen.py already installs telethon when it is missing

- debian 13 and newer ubuntu releases no longer ship neofetch, so the single apt/pacman call failed and installed nothing
- install neofetch on its own and carry on if it is missing; only .sysinfo uses it
- the dpkg/pacman check was inverted: it installed only when a package was already present

- messages.createChat returns InvitedUsers now, so r.chats raised and .create b always failed
- pass the chat object to ExportChatInviteRequest so it resolves without the entity cache

- .alive and inline alive read Repo().active_branch, which raises on a detached HEAD
- fall back to the short commit, or to the main repo link when there is no .git

- every boot after the first calls updater() from ready(), and it raised on a detached HEAD
- fetching a branch that upstream doesn't have, or fetching offline, raised too
- skip the check in those cases; .update reads the branch name the safe way

- on shells whose not-found message the check doesn't recognise, .sysinfo sent an empty render
- the carbon API rejects empty input, so telegram failed with ImageProcessFailedError

- add_snip() stores the snips under the SNIP key, so the "first snip?" check right after it was always false
- the handler was never registered until the next restart; add_handler already skips duplicates

- Poll() now requires a hash, so .poll, .suggest and the trivia quiz all raised TypeError
- correct_answers takes 0-based option indices instead of option bytes
- .suggest passed plain strings where TextWithEntities is required

- youtube-search-python is abandoned and passes proxies= to httpx, which httpx 0.28 removed
- .ytsa/.ytsv and the inline yt search raised TypeError on every query
- search through yt-dlp's flat ytsearch instead; it is already a dependency

- reading im.size leaves PIL's file handle open, so os.remove() failed on Windows
- it also leaked one handle per call elsewhere

- .zip shelled out to the zip binary, which installer.sh never installs
- without it the archive was never written and the upload raised FileNotFoundError
- password-protected zips still need the zip binary; zipfile can't write them

- the python 3.13 rework ran eval code in a bare dict with only event, client and reply
- snippets using asst, udB, ultroid_bot or LOGS raised NameError
- start the dict from the plugin's globals; it stays separate from the module