Retry transient DNS resolution failures

[pudo]

DNS resolution failures are a recurring cause of zavod crawler outages. Pravda should absorb transient DNS failures when a later capture attempt can succeed.

Behavior

• Identify DNS resolution failures separately from other browser and navigation errors.
• Retry them automatically with a small, bounded attempt count and backoff.
• Run each retry in a clean browser context so failed navigation state is not reused.
• Preserve attempt count and the final machine-readable failure cause on the snapshot.
• Log retries with the snapshot ID, hostname, attempt number, and error class.
• Do not retry permanent HTTP or content-validation failures under this policy.

Questions

• Which Chromium/Playwright errors reliably identify DNS failures?
• Should retry limits be global configuration or caller-controlled?
• Does DNS caching in the browser process affect retries?
• Should resolver rotation across system, Quad9, Google, or Cloudflare be supported later? If so, it must be configurable and observable rather than implicit.

Acceptance criteria

• DNS failures have a structured failure cause.
• A transient DNS failure is retried automatically and can produce a successful snapshot.
• Retries stop at the configured bound.
• Exhausted retries produce one inspectable capture result with attempt metadata.
• Tests cover recovery and exhaustion without making real network requests.