GH-6026: Pass reasoning_content back to DeepSeek API in multi-turn conversations

[suryateja-g13]

Summary

• Fixes DeepSeekChatModel.createRequest() to forward reasoning_content from DeepSeekAssistantMessage when rebuilding conversation history for subsequent API calls
• DeepSeek reasoning models (deepseek-reasoner) require the reasoning_content field to be echoed back in ASSISTANT messages in the conversation history — without it, the API returns HTTP 400: "The reasoning_content in the thinking mode must be passed back to the API."
• Previously createRequest() always passed null as reasoning_content in ChatCompletionMessage for ASSISTANT turns, silently dropping the stored value
• The fix checks whether the incoming AssistantMessage is actually a DeepSeekAssistantMessage and, if so, reads getReasoningContent() and passes it forward
• Plain AssistantMessage (non-reasoning models like deepseek-chat) still emit null reasoning_content and are unaffected

Test plan

• ./mvnw -pl models/spring-ai-deepseek test -Dtest="DeepSeekChatCompletionRequestTests" — 3 tests pass (1 existing + 2 new)
• New test reasoningContentIsPassedBackForDeepSeekAssistantMessage verifies the reasoning content is forwarded
• New test nullReasoningContentIsOmittedForPlainAssistantMessage verifies non-reasoning models are unaffected

Fixes: #6026

[suryateja-g13]

The CI failure is the pre-existing flaky test WebClientStreamableHttpTransportErrorHandlingIT.testSessionRecoveryAfter404 in the MCP WebFlux transport integration test suite — unrelated to this PR (tracked in #5563). This test fails intermittently across many PRs that touch unrelated parts of the codebase.