fix: cache key covers synthesis settings, fix silent loudnorm failure, remote sentence silence

Synthesis:

• cache key now includes MODEL, LENGTH_SCALE, SENTENCE_SILENCE, so changing a
  voice or speed no longer serves a WAV built with the old settings; parts are
  joined with length-prefixed values to avoid collisions
• remote mode: piper1-gpl accepts length_w_scale, not noise_w, and ignores the
  latter silently; phoneme width was stuck at its default
• remote mode: sentence_silence is a server start-up argument, not a per-request
  field, so silence is now assembled locally by synthesising each sentence
  separately and concatenating with a gap
• post-processing used a .tmp output name, which ffmpeg rejects with
  "Error initializing the muxer"; the step failed silently, so loudnorm never
  ran and cached audio was clipped at 0.0 dB. Now .tmp.wav with explicit -f wav,
  and a failure is logged instead of swallowed
• loudnorm computes at 192 kHz and leaked that rate into the result, inflating
  files from 186 KB to 2 MB. Sample rate of the source is probed and restored
• LEAD_SILENCE adds a configurable pause before the phrase (default 1.0 s), it is
  part of the cache key
• keep punctuation: the previous [^\p{L}\p{N}\s] filter dropped it all, making
  speech monotonous with no pauses or question intonation
• WAV URL in CLI mode is built from ROOT, so the browser no longer receives a
  filesystem path

Browser and delivery:

• cache buster used filemtime(prepend.php) instead of the script's, so edits to
  piper_tts.js did not change the URL and the old version was served
• wait for readyState 4 / canplaythrough before play(), otherwise the media
  pipeline starts ahead of real time and the start of the phrase is lost
• socket handlers are bound to the captured socket, fixing the null close error
  and CONNECTING sockets being overwritten
• autoplay rejection and load errors are reported instead of swallowed
• WS port comes from the core WEBSOCKETS_PORT constant; status reports the
  websocket separately from piper, since a reachable piper says nothing about
  delivery to a browser tab

Addon renamed to piper_tts_addon: the old class name collided with the module
and produced a fatal "Cannot redeclare class piper_tts". It now supports remote
mode, shares text normalisation, stops leaking dynamic properties, closes its
handles, and falls back between paplay/aplay/ffplay.

Hardening: voice and quality are whitelisted and model paths are confirmed to
stay inside MODELS_DIR, since they reach mkdir/chown/rm; form values are escaped;
install_model and delete_model are refused in remote mode; install/uninstall
survive mysqli exceptions on PHP 8.1+; uname is cached and remote requests use
curl instead of shell.
https://github.com/golnet1/majordomo-piper_tts/com...

feat: SAYTO via SAY_CACHED_READY to terminals, addon reads config from DB
https://github.com/golnet1/majordomo-piper_tts/com...

feat: add SAYTO subscription and destination routing
https://github.com/golnet1/majordomo-piper_tts/com...

help: add nginx+dual-PHP, prepend mechanism, hidden tab WS nuances
https://github.com/golnet1/majordomo-piper_tts/com...

piper_tts.js: close WS on visibilitychange (background tabs)

• WS disconnect on document.hidden, reconnect on visible
• replaces windowActive/tryPlay/localStorage approach
• background tabs no longer receive TTS events at all

install/uninstall: align with Vosk on modules/prepend.php

• install: create shared modules/prepend.php loader if missing
• install: clean up old direct Piper .htaccess line
• install: add auto_prepend_file for modules/prepend.php
• uninstall: check if Vosk is still installed before removing
  shared loader and .htaccess line
• prepend.php: add cache buster via filemtime
  https://github.com/golnet1/majordomo-piper_tts/com...

piper_tts.js: windowActive flag with proper focus target check + localStorage token lock
https://github.com/golnet1/majordomo-piper_tts/com...

help: clarify prepend.php troubleshooting, make section collapsible
https://github.com/golnet1/majordomo-piper_tts/com...

Bootstrap styling: settings and models templates
https://github.com/golnet1/majordomo-piper_tts/com...

prepend: inject script only if not in frame (window.top===window.self)
https://github.com/golnet1/majordomo-piper_tts/com...

cleanup: remove dead code and fix stale references

• Removed unused language constants PIPER_TTS_PLAYER_SETUP,
  PIPER_TTS_AUTO_SETUP, PIPER_TTS_SPA_SETUP from both ru/en files
• Removed dead handlers install_start, install_progress (replaced by install_model)
• Removed creation of unused /tmp/piper-tts directory in install()
• Fixed help.html voice list to match actual 5 voices
• Fixed help.html cache dir default path to match code
• Fixed WS_PORT not being passed to JS (added window.PIPER_WS_PORT
  in piper_tts.html)
  https://github.com/golnet1/majordomo-piper_tts/com...

fix: check prepend.php exists before adding .htaccess entry; prevent system crash on install
https://github.com/golnet1/majordomo-piper_tts/com...