fix: support large Telegram documents

[ponomr]

Summary

• Use configurable Telegram max_document_mb instead of hardcoded 20 MB
• Avoid passing document download/limit errors into the LLM as user text
• Copy local Bot API files directly in local_mode to avoid large download timeouts
• Increase document fetch/download timeouts for large files

Test Plan

• python -m py_compile gateway/platforms/telegram.py
• Restarted bizbot gateway and verified Telegram connection

[teknium1]

Closing — overlaps with the just-merged #28527 (salvage of #25280 'skip-STT audio path + 2GB cap via local Bot API server') and has fork-specific issues:

1. Duplicate mechanism: Both PRs add a configurable document-size cap. feat(telegram): skip-STT audio path + 2GB cap via local Bot API server #25280 uses an automatic 2GB cap when extra.base_url is set (presence of locally-hosted Bot API server is the opt-in). Your PR adds an explicit max_document_mb config. The 2GB-on-base_url approach is simpler — operators who configure a local Bot API already opt in via that.

2. Russian user-facing strings: The new size-limit error messages are in Russian ("Не смогла проверить...", "Документ слишком большой..."). Upstream needs English by default; localization should go through a translation layer.

3. Unrelated bundled commits: Your PR also adds tools/pdf_compress_tool.py (241 LOC) and tools/office_extract_tool.py (319 LOC) which are independent features deserving their own PRs.

The local Bot API file-copy optimization you mention in the summary IS still uncovered — that would be a clean follow-up PR. The English version of "document too large" error messaging would also be welcome. Thanks for the work.