fix: default reasoning_content to empty string for DeepSeek

[amitksingh1490]

Summary

Add a DefaultReasoningContent transformer that ensures the reasoning_content field is always present on messages sent to the DeepSeek provider, falling back to an empty string when no reasoning text exists.

Context

DeepSeek requires reasoning_content to be present on assistant messages even when there is no reasoning text. Without this field, DeepSeek rejects the request. The existing ReasoningContent transformer extracts reasoning text from reasoning_details but leaves reasoning_content as None when there are no reasoning details. This new transformer fills in the gap.

Changes

• Added a new DefaultReasoningContent transformer that sets reasoning_content to an empty string when it is None
• Wired it into the provider pipeline gated exclusively to the DeepSeek provider
• Added unit tests for the transformer itself and the DeepSeek pipeline integration

Key Implementation Details

The new transformer runs after ReasoningContent in the pipeline, so it only fills in the fallback when ReasoningContent (or any earlier transformer) left the field as None. It is gated by is_deepseek_provider() and does not affect any other provider.

Pipeline ordering:

ReasoningContent → DefaultReasoningContent → ...

Testing

# Run the new transformer unit tests
cargo test --package forge_app -- dto::openai::transformers::default_reasoning_content --nocapture

# Run the DeepSeek pipeline integration test
cargo test --package forge_app -- dto::openai::transformers::pipeline::tests::test_deepseek_provider_falls_back_to_empty_reasoning_content_when_none --nocapture

# Run all transformer tests
cargo test --package forge_app -- dto::openai::transformers --nocapture