Signed-off-by: dependabot[bot] <[email protected]>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Owl Bot <gcf-owl-bot[bot]@users.noreply.github.com>

Improved error handling in `google/auth/compute_engine/_metadata.py` to
provide more informative error messages when retries are exhausted.

The logic now handles two distinct failure modes:
1.  **Connection Errors (Exceptions):**
- Exceptions caught during the request (e.g., connection refused) are
stored in `last_exception`.
- If the retry loop exhausts and `last_exception` is set, a
`TransportError` is raised that chains the original exception. This
preserves the stack trace and specific error type for debugging.

2.  **Retryable HTTP Status Codes:**
- If a request completes but returns a retryable status code (e.g., 500,
503), `last_exception` is cleared (set to `None`).
- If the retry loop exhausts due to repeated retryable status codes, a
`TransportError` is raised that includes the `response.status` and the
response body (`error_details`). This ensures the specific server error
message is visible.

This addresses feedback from PR 1637 by ensuring that "Ran out of
retries" scenarios provide the most relevant context: either the
underlying exception or the final HTTP error response.

---
*PR created automatically by Jules for task
[6226991344715693997](https://jules.google.com/task/6226991344715693997)
started by @chalmerlowe*

---------

Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>
Co-authored-by: Chalmer Lowe <[email protected]>