Why?

Ruby's net/http can raise Net::HTTPFatalError for 5xx responses (link) before stripe-ruby converts the response into a Stripe::StripeError. Because our retry logic only retried known network exceptions and Stripe::StripeError instances, these transient 5xx failures were not retried.

This change makes retry behavior consistent for that code path so temporary 5xx failures surfaced as Net::HTTPFatalError are retried like other transient server-side failures.

What?

• treat Net::HTTPFatalError as retryable in APIRequestor.should_retry?
• map Net::HTTPFatalError through NETWORK_ERROR_MESSAGES_MAP so exhausted retries raise Stripe::APIConnectionError
• add tests covering both the retry classification and persistent-failure retry behavior

Changelog

• Fixes a bug where Net::HTTPFatalError errors were being thrown and not retried (unlike other 5xx responses from the Stripe API)