Fix the API grounding check for agent code

An agent writing code against a third-party SDK keeps calling methods that do not exist. Each "fix" invents another plausible name. Instead of a third prompt, you check its calls against the SDK's real method list, taken from the type definitions.

An assistant wrote ground_calls_against_api(api_methods, called_methods). It is in the editor. It has a bug. Find it and fix it.

Input and output:

• api_methods: the real methods of the SDK client.
• called_methods: the methods the agent's code calls, in call order.
• Return one call=method or call=missing string per call that is not grounded, in first call order.

Rules the tests check:

• Trim whitespace from both ends of every name. Ignore names that are then empty. Handle each called name once, at its first call.
• A call is grounded only when it equals a real method exactly, case included. Report nothing for it.
• The shape of a name is the name with every _ and - removed and the letters A to Z turned into lowercase. Nothing else changes.
• A call that is not grounded but has the same shape as a real method, like sendMessage for send_message, is reported as call=method. If several real methods have that shape, use the smallest name, compared as plain strings.
• Any other call is reported as call=missing.