fix(dingtalk): drain inbound background tasks

[yu-xin-c]

Summary

• observe DingTalk inbound task completion instead of only discarding task handles
• retrieve and log unexpected task failures with their traceback
• treat cancellation as normal and await every tracked task during channel shutdown

Why

The DingTalk stream callback forwards accepted messages through background tasks so it can acknowledge the platform promptly. The previous completion callback removed each task from the registry without retrieving its result, turning message-processing failures into detached Task exception was never retrieved warnings.

Shutdown also cancelled and immediately cleared the registry without awaiting the tasks. DingTalkChannel.stop() could therefore return while inbound work was still cancelling.

The shared completion callback now owns terminal observation, while shutdown snapshots, cancels, and gathers the same tracked task set before returning.

Validation

• regression on unmodified main: 2 failed (Task exception was never retrieved; task still cancelling after stop())
• focused lifecycle regressions: 2 passed
• full DingTalk test module: 39 passed
• pytest -q -n auto: 6151 passed, 23 skipped
• ruff check on changed files
• basedpyright nanobot/channels/dingtalk/runtime.py: 0 errors
• git diff --check

Closes #5463

[Re-bin]

LGTM. DingTalk inbound tasks now have observed terminal states, shutdown waits for tracked cancellations, and late file/rich-text handlers cannot create new work after shutdown begins.

Focused lifecycle regressions cover failure, cancellation, and the download/stop race, and all required checks are green on the current head based on the latest main.