fix(notion-client): wait out rate-limit cooldowns instead of backing off

The initial retry policy backed off from 1s (1s -> 2s -> 4s, 30s cap), which
cannot succeed against this API: once a client trips a rate limit, Notion
applies a fixed cooldown, so every quick retry lands inside the same window
and just burns the retry budget before surfacing the failure.

- On 429, wait out the cooldown (60s plus jitter) rather than backing off
  from a short delay.
- Raise the max delay from 30s to 60s.
- Use full jitter (uniform over [0, exponential]) for transient errors, which
  decorrelates retry waves better than a fixed fraction of the window.
- Narrow the retried status codes to [408, 429, 500, 502, 503, 504], dropping
  409 and 425. Private pages return 400 and now fail fast instead of
  consuming retries. 500 is kept because ofetch synthesizes it for
  transport-level failures.
- Classify on response status before touching the body, and tolerate non-JSON
  error bodies. Read the optional clientData.retryAfter hint as a fallback,
  since these responses generally carry no Retry-After header.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>