emrg: the feishu template gives a round no way to tell an empty inbox from an unreadable one #2621

emrg/server/feishu_prompt.md §1 ends with one sentence about empty rounds: "A round that reads nothing must say so." The template never says how a round is supposed to know that the round it just had was empty rather than blind.

That distinction is not academic. A listing (im +chat-messages-list) that comes back with zero messages is the same answer in four different situations:

1. the conversation really has nothing new;
2. the identity resolved to the wrong account, so the call read a conversation that is not the host's;
3. a token valid under one HOME reads missing under another — the template already says this symptom exists (§1, the missing identity bullet) but treats it only as a blocked item;
4. the index is lagging, which the template names for im +messages-search only, and only to say that the search "can never establish that nothing came back".

The round therefore has a report it is instructed to make ("nothing arrived") and no instrument that earns it. Reported as-is, that sentence is indistinguish­able from a checked one, and this task's entire product is the difference between a quiet inbox and an unread one.

What the fix must satisfy

• The rule belongs in §1 of emrg/server/feishu_prompt.md, which owns reading — not appended as a new section, and not duplicated in §4's report requirements.
• It states the general method: before an empty answer is reported, one conversation known to hold messages is read, and only a call that returns them shows the interface works — so that "nothing new" then means nothing arrived rather than nothing was readable.
• How often that check runs is a per-installation tuning decision and stays out of the template.
• A test names the requirement, both in the template file and in the render that a round is actually sent: the template is Jinja2, so a construct can keep every character in the file and send none of them.