Interrupted turns now report a resume_reason across streaming, SDK results, and resumed prompts

What

• Streaming partial-message frames can now carry a resume_reason field explaining why the current turn is an automatic re-run of one interrupted by a worker restart: host_draining, checkpoint_restore, container_recreated, or a fallback interrupted_turn. This ties to the CLAUDE_CODE_RESUME_INTERRUPTED_TURN / CLAUDE_CODE_RESUME_REASON mechanism.
• The SDK/headless turn-result schema gains the same optional resume_reason string, plus an optional result_index integer, in both the streamed and final result message shapes.
• When Claude Code auto-resumes an interrupted turn at session restart, the reconstructed prompt now includes a resumeReason and, when available, a resumedTurnUserMessageUuid pointing back to the original interrupted turn's user message.

Why This makes it clear, both in the UI/streaming output and in SDK results, why a turn was automatically re-run after an interruption (such as the worker host draining or a container being recreated), and links the resumed turn back to the original one it replaced.