Summary

Fixes #79605.

Telegram partial preview mode now starts the draft preview immediately instead of applying the block-streaming first-message debounce. That preserves the debounce for block previews, while allowing direct model partial text to create the live Telegram preview before final delivery.

The regression test uses the real Telegram draft stream transport from the dispatcher test, emits a short direct partial (Hi), verifies the draft transport sends the preview before final delivery, then verifies the final answer edits that preview instead of falling back to the normal final-send path.

Root Cause

dispatchTelegramMessage passed DRAFT_MIN_INITIAL_CHARS to the draft stream for both partial and block modes. Short direct model partials in partial mode could stay below the 30-character initial threshold, so no visible Telegram preview was created even though partial callbacks reached the lane.

Real behavior proof

From extensions/telegram/src/bot-message-dispatch.test.ts:

• A direct model partial Hi is emitted through replyOptions.onPartialReply.
• The real Telegram draft transport calls bot.api.sendMessage(123, "Hi", ...) before any final delivery.
• The final answer Hi there calls bot.api.editMessageText(123, 777, "Hi there", ...).
• deliverReplies is not called, proving the turn stayed on the preview/edit path instead of sending only a final Telegram message.

Validation

• pnpm test extensions/telegram/src/bot-message-dispatch.test.ts extensions/telegram/src/draft-stream.test.ts src/plugin-sdk/channel-streaming.test.ts
• pnpm exec oxfmt --check --threads=1 extensions/telegram/src/bot-message-dispatch.ts extensions/telegram/src/bot-message-dispatch.test.ts
• git diff --check
• pnpm check:changed --base upstream/main