Add transient-network retry with exponential backoff

- bls_client/retry.py: with_retries() retries timeouts/connection drops and
  429/5xx with exponential backoff; honors Retry-After; never retries 4xx or
  BLSQuotaError (those won't fix themselves)
- BLSClient(retries=2, backoff=0.5) wraps the data POST and discovery GETs;
  qcew CSV fetch wrapped too. retries=0 disables.
- tests/test_retry.py: 6 offline tests (injectable sleep, no real delays)
- README: retry behavior documented; drop the now-resolved limitation

Offline suite 125 passing; live smoke green.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

bls_client/client.py

@@ -12,6 +12,7 @@ from datetime import datetime
 
 from .cache import FileCache
 from .errors import BLSError, BLSRequestError, BLSQuotaError
+from .retry import with_retries
 
 
 class BLSClient:
@@ -19,7 +20,8 @@ class BLSClient:
     MAX_SERIES_PER_CALL = 50
     MAX_YEARS_PER_CALL  = 20
 
-    def __init__(self, api_key: str, cache=False, cache_dir=None, cache_ttl=86400):
+    def __init__(self, api_key: str, cache=False, cache_dir=None, cache_ttl=86400,
+                 retries=2, backoff=0.5):
         """
         Args:
             api_key:   BLS registration key.
@@ -27,6 +29,9 @@ class BLSClient:
                        or pass a custom cache object with get(key)/set(key, value).
             cache_dir: Override the default cache directory.
             cache_ttl: Cache entry lifetime in seconds (default 1 day).
+            retries:   retries after the first attempt on transient network errors
+                       (timeouts, connection drops, 429/5xx). 0 disables retrying.
+            backoff:   base exponential-backoff interval in seconds.
         """
         self.api_key = api_key
         if cache is True:
@@ -35,8 +40,15 @@ class BLSClient:
             self.cache = cache            # caller-supplied cache object
         else:
             self.cache = None
+        self.retries = retries
+        self.backoff = backoff
         self.queries_used = 0             # network POSTs spent this session (cache hits don't count)
 
+    def _get(self, url, **kwargs):
+        """GET with retry/backoff."""
+        return with_retries(lambda: requests.get(url, **kwargs),
+                            retries=self.retries, backoff=self.backoff)
+
     # ------------------------------------------------------------------
     # Core fetch
     # ------------------------------------------------------------------
@@ -97,7 +109,9 @@ class BLSClient:
 
     def _fetch_batch(self, payload: dict) -> dict:
         """Execute one (uncached) API call and return {series_id: {data, catalog}}."""
-        r = requests.post(f"{self.BASE_URL}/timeseries/data/", json=payload, timeout=30)
+        url = f"{self.BASE_URL}/timeseries/data/"
+        r = with_retries(lambda: requests.post(url, json=payload, timeout=30),
+                         retries=self.retries, backoff=self.backoff)
         r.raise_for_status()
         self.queries_used += 1
         body = r.json()
@@ -155,13 +169,13 @@ class BLSClient:
     # ------------------------------------------------------------------
     def surveys(self) -> list[dict]:
         """Return all BLS surveys with abbreviation and name."""
-        r = requests.get(f"{self.BASE_URL}/surveys", timeout=15)
+        r = self._get(f"{self.BASE_URL}/surveys", timeout=15)
         r.raise_for_status()
         return r.json()["Results"]["survey"]
 
     def popular(self, survey: str) -> list[str]:
         """Return popular series IDs for a given survey abbreviation."""
-        r = requests.get(
+        r = self._get(
             f"{self.BASE_URL}/timeseries/popular",
             params={"survey": survey},
             timeout=15,