Expose token usage from openAIResponsesAdapter

[rabisg]

Is your feature request related to a problem? Please describe.
openAIResponsesAdapter() currently ignores response.completed, including response.usage.

Describe the solution you'd like
Could the adapter expose an optional callback?

openAIResponsesAdapter({
onUsage: (usage) => {
// input_tokens, output_tokens, total_tokens
},
});
This would let applications display token usage or estimated remaining context without implementing a custom stream adapter.

Describe alternatives you've considered
Local adapter that wraps/forks openAIResponseAdapter

Additional context
This would help implement a context available/used counter in the composer

[justonemorenight]

I've submitted a PR in #1147 to add the onUsage callback option to openAIResponsesAdapter.