Preserve OpenAI audio transcription output

[Hamed Rabah (hamedrabah)]

Closes #131.

Summary

• recognize OpenAI transcription and translation responses by their text field
• preserve the complete response object in braintrust.output_json, including verbose segments and words
• keep existing choices/output precedence unchanged

Testing

• ./gradlew :braintrust-sdk:test --tests dev.braintrust.instrumentation.InstrumentationSemConvTest
• ./gradlew check

[Hamed Rabah (hamedrabah)]

Abhijeet Prasad (@AbhiPrasad) ready for review when you have a moment. The full local CI-equivalent suite passes (./gradlew check: 297 tasks). GitHub currently reports no checks for this fork PR, so Actions may need maintainer approval to start.

[Andrew Kent (realark)]

thanks for sending this Hamed Rabah (@hamedrabah) , i'll get this out in the next release