XTester v.0.0.72

Telegram control of simulations, smoother AI Copilot chat, Research Lab import, project-metadata sync.

Telegram:

• A simulation can now pause on Telegram.Ask, wait for a real inline-button answer in wall-clock
  time, and automatically resume the accelerated run from the same simulation point.
• Launch profiles can instead auto-select button 1, 2, or 3, or leave a question unanswered,
  so deterministic backtests can pass Telegram decisions without pausing.
• Telegram decisions are restricted to the originating chat/message and authorized user IDs;
  callback tokens are session-scoped so stale buttons cannot affect a later run.

AI Copilot:

• Chat rendering goes through a sequential render queue: streaming deltas are merged instead of
  piling up, and the latest turn updates atomically without page reloads — long streamed answers
  no longer freeze or flicker the chat.
• The chat input box keeps its chosen height across window resizes and restarts.

Research Lab:

• Importing a project into the Lab builds the spec from the real strategy code, and fresh backtest
  metrics (when available) are included in the context; the spec can be edited in the spec window
  before the research starts.
• The research title is generated by the LLM from the finished spec, with a deterministic fallback.
• Local reasoning models (LM Studio): thinking output is reliably stripped from answers.

Project sync:

• Project name, description, and version are merged three-way with the on-disk project, and window
  captions / explorer labels update when metadata arrives from disk — external edits no longer
  clobber each other.

Docs & localization:

• Help updated (Launcher, Projects, Research Lab, window reference) in Russian and English;
  continued English localization of new UI messages.