A short sample text used by the smoke E2E tests.

The quick brown fox jumps over the lazy dog. This sentence is intentionally
boring so that any LLM-style summary preset has unambiguous content to work
with. The text continues for several short paragraphs to give the chunker
something to size, but never exceeds a few hundred tokens so the test stays
fast and the OpenAI mock can return a deterministic response.

Section: Why this fixture exists.

Smoke E2E tests answer one question: does `unread` survive a full
analyze round-trip on a fresh install with a real (but small) input file?
Without this, every other test could pass while the orchestrator throws
on a missing module, a stale schema migration, or a broken format-check.
The fixture is text-only so no ffmpeg / Whisper / vision dependency
sneaks into the test path.

Section: What we expect.

The smoke run mocks the AI provider. A real run would hit OpenAI / Anthropic
/ Google. The point is to confirm the wiring (CLI → command → settings →
provider dispatch → formatter → cache write → report file) holds together
end-to-end, not to validate the model output itself.
