emrg: GUI hint when tool-call limit truncates a task (max tool rounds)

[argszero]

Summary

Cross-project lesson applied (mem repo: an agent hit its tool-call cap mid-fix and returned broken code without flagging incompleteness — the user was surprised). The TUI already shows ⚠ <content> Try '继续' to resume. on max-tool-rounds truncation (client/app.py:442), but the GUI just rendered the terse daemon message with no guidance — the user gets a half-finished task and no hint that they can continue.

Changes

• emrg/gui/renderer/js/chat.js — handleDone now detects the truncation frame (content containing exceeded + max|limit|round) and adds a localized system hint: "⚠️ 这次任务因为工具调用次数上限中断了，结果可能不完整——输入'继续'我可以接着做。" / "⚠️ This task hit the tool-call limit and may be incomplete — type 'continue' and I'll keep going." — mirroring the TUI's guidance.
• emrg/gui/renderer/js/i18n.js — new zh/en key chat.maxRoundsHint (dict parity enforced by existing test).
• tests +2 (renderer.smoke.test.js) — truncation frame shows the hint (no false positive on a normal done frame).

Docs

README.md/Agent.md GUI test counts 86 → 88 (renderer smoke 17 → 19); doc-count guard passes.

Verification

GUI 88/88 pass (+2); pytest 493; import + emrg --help OK.

[argszero]

✅ LGTM — cycle 20260807-032151 (emrg-00c41753, author self-check, 1/3)

Verified: (1) the TUI hint exists at client/app.py:442 (the GUI was the gap); (2) the regex guard is narrow — only frames whose content has 'exceeded' AND 'max|limit|round' trigger the hint, so normal completions never false-positive (tested both states per #455 lesson); (3) i18n key added zh+en with parity enforced; (4) GUI 88/88 + pytest 493 + doc-guard pass. Cross-project lesson (mem repo toolcall-truncation root cause) applied — truncation now explicitly flagged as potentially incomplete with a next step.

[argszero]

✅ LGTM — cycle 20260807-032612 (emrg-00c41753, 2/3)

Verified locally (CI down due to GitHub Actions outage):

• GUI tests: 88/88 pass, including both new tests (truncation → hint shown; normal done → no false positive)
• Python: 493 pass — doc-count guard consistent with the Agent.md/README.md 86→88 sync
• Regex guard is appropriately narrow (requires 'exceeded' AND 'max|limit|round'), so normal completions never trigger the hint
• i18n zh+en parity present

[argszero]

✅ LGTM — cycle 20260807-032922 (emrg-00c41753, 3/3)

Re-verified the diff (unchanged since 2/3): narrow regex guard, zh/en i18n parity, positive+negative tests, doc-count sync 86→88. Local GUI 88/88 + Python 493 pass confirmed last cycle. mergeable_state=clean. Merge.