fix(simulation): retry temporary provider proxy failures

src/k1link/simulation/gaussian_pipeline_gateway.py

@@ -36,6 +36,7 @@ IMAGE_DIGEST_PATTERN: Final = re.compile(r"^sha256:[a-f0-9]{64}$")
 DEFAULT_CHUNK_BYTES: Final = 8 * 1024 * 1024
 MAX_JSON_RESPONSE_BYTES: Final = 32 * 1024 * 1024
 MAX_RETRIES: Final = 3
+RETRYABLE_PROVIDER_STATUS_CODES: Final = {502, 503, 504}
 DEFAULT_INGEST_TIMEOUT_SECONDS: Final = 30 * 60.0
 
 
@@ -899,7 +900,10 @@ def _validate_runtime_provenance(document: Mapping[str, object]) -> None:
 
 
 def _unavailable(message: str, error: httpx.HTTPError) -> GaussianPipelineGatewayError:
-    if isinstance(error, httpx.TransportError):
+    if isinstance(error, httpx.TransportError) or (
+        isinstance(error, httpx.HTTPStatusError)
+        and error.response.status_code in RETRYABLE_PROVIDER_STATUS_CODES
+    ):
         return GaussianPipelineUnavailableError(message)
     return GaussianPipelineGatewayError(message)
 
@@ -922,4 +926,6 @@ def _provider_rejection(response: httpx.Response) -> GaussianPipelineGatewayErro
     message = f"Gaussian provider rejected request (HTTP {response.status_code})"
     if detail is not None:
         message = f"{message}: {detail}"
+    if response.status_code in RETRYABLE_PROVIDER_STATUS_CODES:
+        return GaussianPipelineUnavailableError(message)
     return GaussianPipelineGatewayError(message)