test(waterdata): retry every transient error in the flaky-rerun filter (#325)

The suite already retries transient HTTP failures (flaky's `only_rerun`), but
the patterns missed two kinds, so a transient upstream 502 failed CI instead of
retrying:

- a direct 5xx is raised as `ServiceUnavailable`, and
- a chunked request wraps a transient 429/5xx as `QuotaExhausted` /
  `ServiceInterrupted`.

Add patterns for both. Verified they retry these transient errors but not
deterministic ones (e.g. a 404 or an assertion failure).

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

Author: thodson-usgs

tests/waterdata_test.py

@@ -53,7 +53,14 @@
     reruns=2,
     reruns_delay=5,
     only_rerun=[
-        r"(?:RateLimited|RuntimeError):\s*(?:429|5\d\d):",  # _raise_for_non_200 output
+        # Transient HTTP errors (429 / 5xx) on the direct path: RateLimited /
+        # ServiceUnavailable carry a "<status>: ..." message (the RuntimeError
+        # shape is kept for any legacy call site).
+        r"(?:RateLimited|ServiceUnavailable|RuntimeError):\s*(?:429|5\d\d):",
+        # The chunked fan-out wraps a transient sub-request as a ChunkInterrupted
+        # subclass (QuotaExhausted for 429, ServiceInterrupted for 5xx), whose
+        # message has no leading status token.
+        r"(?:QuotaExhausted|ServiceInterrupted):",
         r"Connect(ion)?Error",  # requests' ConnectionError + httpx' ConnectError
         r"ReadTimeout|ConnectTimeout|Timeout",
     ],