fix: restore continuation when truncated output is empty

[xbx787]

What changed

• Added an empty-output guard to callLlmWithContinuation.
• When the initial response is truncated without visible content, retry the original task once with thinkingMode: "fast".
• Allow bounded continuation after an empty retry result, using a dedicated prompt when there is no assistant text to prefill.
• Preserve the existing non-empty continuation path and graceful failure behavior.

Why

Reasoning models can consume the entire max_tokens budget internally and return finishReason: "length" with empty message.content. The previous text.trim() loop guard prevented any continuation request, producing an empty final briefing. This fixes issue #2.

Fixes #2

Impact

• Normal non-empty truncated responses keep the original options object and continuation behavior.
• Empty truncated responses get one fast-mode retry, followed by at most the configured continuation count.
• Retry and continuation failures preserve the best available output instead of failing the whole task.
• No generated build artifacts or version files were changed.

Validation

• npm test — 24 test files, 264 tests passed.
• npm run typecheck:core — passed.
• git diff --check — passed.