Fix streaming aggregation issues

[Stephen Belanger (Qard)]

Summary

• fix streaming usage alias handling so maps to prompt tokens and maps to completion tokens
• aggregate streamed tool calls across chat completion deltas into final output
• preserve detailed streaming usage metrics including token detail breakdowns and prediction token counts
• add focused tests covering aliases, tool call aggregation, and detailed usage extraction

Testing

running 7 tests
test stream::tests::skips_keep_alive_markers ... ok
test stream::tests::aggregates_detailed_usage_from_final_chunk ... ok
test stream::tests::aggregates_content_from_streaming_values ... ok
test stream::tests::aggregates_usage_with_input_output_token_aliases ... ok
test stream::tests::aggregates_usage_from_final_chunk ... ok
test stream::tests::caches_finalized_result ... ok
test stream::tests::aggregates_tool_calls_from_streaming_values ... ok

test result: ok. 7 passed; 0 failed; 0 ignored; 0 measured; 139 filtered out; finished in 0.00s

Fixes #41
Fixes #30
Fixes #32