### What happened?

### Bug Description When a client disconnects during a non-streaming Bedrock API call, LiteLLM fails to accurately track the full cost of the request, even though AWS Bedrock continues processing and charges for the complete request.

### Expected Behavior - LiteLLM should track the full cost of all Bedrock requests, regardless of client connection status - Spend tracking should match AWS billing accurately - Cost logs should include both input and output token costs

### Actual Behavior - Pre-call logging captures request metadata and estimated input costs before sending to AWS - When clients disconnect during processing, post-call logging is skipped - AWS continues processing and charges for the full request (input + output tokens) - LiteLLM only records partial costs (input tokens only), creating discrepancy with AWS billing

### Technical Root Cause In the Bedrock provider implementation: 1. `pre_call()` logging occurs at `litellm/llms/custom_httpx/llm_http_handler.py:352` BEFORE the HTTP request to AWS 2. `post_call()` logging occurs after AWS response is received in the provider's `make_call()` function 3. Client disconnects trigger task cancellation, prev…