fix(news): keep briefing generation green when the Copilot CLI fails

[Copilot]

The scheduled News job "Generate briefing state" has failed on every run since 10 Sep with task: Failed to run task "pipeline:generate": exit status 1 (step exit code 201 = go-task's CodeTaskRunError), so no state has been persisted and the site is stale.

Root cause

pipeline:generate aborts whenever the Copilot CLI subprocess fails. Running the pipeline against the real restored news-state release asset with AI_PROVIDER=fake succeeds, which isolates the failure to the Copilot CLI path; no functional commits landed in the failing window (Renovate bumps only).

Of the three Copilot CLI providers in src/wazzup/ai.py, only CopilotCliTransparencyReportProvider tolerated a CLI outage. The curator and writer still raised on a non-zero exit code, a missing output file, or unparsable JSON — killing the run before any state was persisted. Reproduced with a stub copilot that exits 1: the pipeline exits 1 before this change, 0 after.

Changes

• src/wazzup/ai.py — curation and summary providers now degrade to their deterministic Fake* counterparts on CLI failure, recording provider.type: copilot-cli-fallback, fallbackFrom, and the CLI stdout/stderr as fallbackReason. This is the contract already used for invalid payloads and for the transparency report; the exception tuples of all three are now aligned.
• warn_copilot_fallback — emits a ::warning:: per fallback so the degradation and the underlying CLI error surface in the workflow log instead of failing silently. Written to stderr to keep the pipeline's JSON stdout clean.
• tests/test_ai.py — two tests mirroring test_copilot_cli_transparency_falls_back_on_runtime_failure for the curation and summary providers.
• docs/github-actions.md — operational learning note.

With a CLI that exits non-zero, a run now looks like:

::warning::Copilot CLI curation fell back to deterministic output: Copilot CLI curation failed with exit code 1. ... stderr: premium request quota exhausted
::warning::Copilot CLI summary fell back to deterministic output: ...
::warning::Copilot CLI transparency report fell back to deterministic output: ...

and the published briefing carries the reason:

{
  "type": "copilot-cli-fallback",
  "fallbackFrom": "copilot-cli",
  "fallbackReason": "Copilot CLI failed with exit code 1. ...\nstderr: premium request quota exhausted"
}

Note

api.github.com is unreachable from the agent sandbox, so only the job annotations were retrievable — the precise Copilot CLI error (quota, model availability, or a CLI regression) is still unknown. The next run will print it as a warning and persist it in fallbackReason, which should pinpoint it. Until then briefings refresh with deterministic, non-AI summaries rather than not refreshing at all.