Refactors error handling and retry logic to be more robust and consistent.

There were a number of issues related to this which are now addressed. Tried to simplify the flow and make the main data retrieval route more readable.

Key changes:

• Retry all 5xx errors, not just 502 (fixes API request returned HTTP 500: Internal Server Error #140)
• Retry network errors including URLError, socket.error, and IncompleteRead (fixes Retry API requests which failed due to whatever reasons #110)
• Retry JSON parse errors on HTTP 200 responses (transient corruption)
• Add exponential backoff with jitter (1s base, 120s max, 10% jitter)
• Respect retry-after and rate limit headers per GitHub API requirements
• Consolidate retry logic into make_request_with_retry() wrapper
• Improve error visibility with clear logging before failures
• Remove dead code from 2016 that was intentionally disabled

Fixes #140, #110, #138

#138 suggests adding a --continue-on-error behavior, this is more part of a per-repo solution that relates to #244

Retry behavior

Logging improvements

Clear visibility into retry behavior and failures:

Dead code removed

• _request_http_error() - was dead code, errors list intentionally disabled in 2016 (commit 1e5a904) to fix Quits after resuming from rate limit sleep #29
• _request_url_error() - same as above
• errors list parameter threading - never used

Parameter rename

The retrieve_data() function parameter single_request has been renamed to paginated with inverted logic:

• Old: single_request=True meant "don't paginate"
• New: paginated=False means "don't paginate"

This is more intuitive (positive framing). All internal call sites have been updated.

Tests

New test added to exercise retries. Live tested against a number of repos.