Why?

Platform engineers need to swap the HTTP transport in StripeClient — e.g., routing requests through an internal gRPC API gateway instead of Net::HTTP, or wanting to use a different http library.

Today, StripeClient hardcodes APIRequestor.new(config) with no extension point, and the actual HTTP call is buried inside a private method with no override seam. This PR adds the minimal scaffolding to make the transport pluggable without forking the client.

What?

• Added a requestor: kwarg to StripeClient#initialize that accepts a factory callable (StripeConfiguration) -> APIRequestor, with validation that the returned instance is an APIRequestor
• Extracted a protected send_request method on APIRequestor as the subclass override point for transport, replacing the inline connection_manager.execute_request call
• Wrapped execute_request and execute_request_stream with an active_requestor save/restore pattern so that nested operations (generated resource methods, save, etc.) resolve the custom requestor via thread-local lookup
• Added tests covering factory callable, validation, transport override end-to-end, and active_requestor propagation/restoration
• Updated Sorbet type signatures in rbi/stripe/stripe_client.rbi

See Also

This is PR 1 of 2. A follow-up PR will extract retry/error seams (should_retry?, sleep_time, network_error_map) as overridable instance methods.