fix: retry transient failures when fetching the docs index source

[ChiragAgg5k]

Why

The daily Docs workflow (run 34728251207) and the integration job on main (run 34774829942) both failed on 2026-09-13 with an httpx.ConnectError raised during the TLS handshake of one page fetch against appwrite.io. A full index build issues one request per manifest entry (684 today), and fetch_page had no retry, so a single transient failure aborted the whole rebuild. The CI rerun passed and the full fetch completes cleanly locally, which confirms the failure is transient.

What

• Add get_with_retry in docs_source.py, used by both the manifest fetch and every page fetch.
• Retry httpx.TransportError (connect, read, protocol errors) and 429/500/502/503/504 responses up to 4 attempts with exponential backoff starting at 0.5s.
• 404 and every other status are returned immediately, so unpublished pages still skip without delay and non-retryable errors still surface through raise_for_status.

Validation

• ruff check, black --check, pyright clean.
• 258 unit tests pass.

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge with no outstanding correctness or repository-rule violations.

Summary

• Honors bounded Retry-After values and applies jittered exponential fallback delays.
• Treats non-finite Retry-After values such as NaN and infinity as malformed.
• Preserves immediate handling of non-retryable responses, including 404 pages.

_{Reviews (3) · Last reviewed commit: "fix: reject non-finite Retry-After value..."}