Responses file/reasoning fixes and more test cases

[choochootrain]

Defined a few more cases we might want to care about for multi-turn, reasoning, and multi-modal for all providers and fix the discovered issues:

responses:

• Fix file and reasoning part handling
• Modify responses tests to roundrip OutputItem -> Message -> OutputItem instead of going through InputItem - this is consistent with other provider tests, we can reconsider if necessary

chat completions:

• Fix file part handling

test improvements:

• Use colorized diffs for all test runner failure messages
• Removed duplicate toolCallRequest case from advanced.ts
• Moved reasoningWithOutput to simple.ts