fix(free-code provider): increase retry count and backoff window for 429s

Rate-limit retries were too aggressive (3 attempts, max 8s wait). Since
the free-code proxy's x-app: cli fix may not be deployed everywhere yet,
the repowise client now retries up to 6 times with exponential backoff up
to 60 seconds, covering sustained rate-limit windows.

Also adds before_sleep_log so retry attempts are visible in logs.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: dmikushin

packages/core/src/repowise/core/providers/llm/free_code.py

@@ -29,8 +29,10 @@
     retry_if_exception_type,
     stop_after_attempt,
     wait_exponential_jitter,
+    before_sleep_log,
     RetryError,
 )
+import logging
 
 from repowise.core.providers.llm.base import (
     BaseProvider,
@@ -46,9 +48,13 @@
 
 log = structlog.get_logger(__name__)
 
-_MAX_RETRIES = 3
-_MIN_WAIT = 1.0
-_MAX_WAIT = 8.0
+# More aggressive retry settings than other providers: the free-code proxy
+# inherits Claude Code's rate-limit tier only when x-app: cli is present
+# (fixed in free-code serve). Until old binaries are updated, the client
+# retries longer to absorb transient 429s.
+_MAX_RETRIES = 6
+_MIN_WAIT = 2.0
+_MAX_WAIT = 60.0  # rate-limit backoffs can be long; wait up to 1 minute
 
 _DEFAULT_BASE_URL = "http://localhost:3180"
 
@@ -136,6 +142,7 @@ async def generate(
         retry=retry_if_exception_type(ProviderError),
         stop=stop_after_attempt(_MAX_RETRIES),
         wait=wait_exponential_jitter(initial=_MIN_WAIT, max=_MAX_WAIT),
+        before_sleep=before_sleep_log(logging.getLogger(__name__), logging.WARNING),
         reraise=True,
     )
     async def _generate_with_retry(