Show an empty state when the AI finds no tasks, instead of an error

[Guotai812]

Summary

The backend expressed "the user said nothing task-shaped" by throwing (AiErrorCode.NoTasksExtracted), so a valid empty outcome arrived on the error channel: red toast, transcript wiped, blank screen. Users could not tell a mis-hearing from a genuinely task-free sentence, or keep what they said.

• Backend: an empty extraction now returns a normal AiGenerateMessage with isEmptyResult. Genuine faults (cancel, 429, content filter, unhandled) still throw.
• Why a flag, not empty lists: the extracted* lists are cumulative for the whole session, so a barren turn can still return a non-empty snapshot. ToolCallCount is the only per-turn signal.
• Mobile: the sheet shows what was heard plus a "Save as note" action; drafts from earlier turns stay put.
• Empty results stop counting toward ai_task_generation_failed, which they were 59% of.

Backend and app must ship together: older app versions get no empty-state feedback.

Release note

When the AI doesn't find any to-dos in what you said, you now see what it heard and can save it as a note, instead of a red error message.

Status:

• User-facing — announce it
• Beta / partial — announce, but tagged as beta
• Hidden in production (feature-flagged / not enabled for users) — don't announce
• Internal only (refactor / infra / tests / CI / deps) — don't announce

[github-actions]

Test Results

98 tests   98 ✅  9s ⏱️
 1 suites   0 💤
 1 files     0 ❌

Results for commit 5743d10.

♻️ This comment has been updated with latest results.

[github-actions]

Package	Line Rate	Branch Rate	Complexity	Health
BlotzTask	4%	32%	2593	❌
BlotzTask	4%	32%	2593	❌
Summary	4% (4136 / 104766)	32% (850 / 2660)	5186	❌