feat: capture Telegram reaction feedback

[msheldyakov]

Summary

• Add an opt-in Telegram inbound reaction feedback handler (TELEGRAM_REACTION_FEEDBACK / telegram.reaction_feedback).
• Add a bounded local reaction feedback store under $HERMES_HOME/state with a sent-message index and append-only normalized JSONL events.
• Index Hermes-sent Telegram message ids from rich, legacy, edit, and overflow send paths so user reactions can be tied back to known assistant messages.
• Store route/session metadata plus content hashes only; do not store raw assistant text or raw actor user ids; do not auto-apply feedback to prompts, memory, or skills.

Test Plan

• python3 -m pytest tests/gateway/test_reaction_feedback_store.py tests/gateway/test_telegram_reactions.py -q -o 'addopts='
• python3 -m pytest tests/gateway/test_reaction_feedback_store.py tests/gateway/test_telegram_reactions.py tests/gateway/test_telegram_rich_messages.py tests/gateway/test_telegram_send_path_health.py -q -o 'addopts='
• python3 -m pytest tests/gateway/test_telegram_thread_fallback.py -q -o 'addopts='
• python3 -m py_compile gateway/reaction_feedback.py gateway/platforms/telegram.py gateway/config.py tests/gateway/test_reaction_feedback_store.py tests/gateway/test_telegram_reactions.py
• git diff --check

[alt-glitch]

Related: #23731 (zetxek) already implements Telegram reaction-feedback capture for the same feature requests (#18408 / #27438). Different mechanism — this PR uses an opt-in bounded $HERMES_HOME/state store with a sent-message index and content-hash-only events; #23731 logs to feedback.jsonl + memory/feedback-log.md. Competing approaches, not a duplicate — maintainer picks one.