Fix DeepSeek reasoning_content echo on fresh tool calls

[punkcanyang]

Summary:

• persist reasoning_content="" on fresh assistant tool-call turns for DeepSeek/Kimi-style thinking providers
• keep poisoned-history replay fallback in _copy_reasoning_content_for_api
• add a regression test covering _build_assistant_message for DeepSeek tool-call replies without explicit reasoning_content

Root cause:
_build_assistant_message() checked the DeepSeek fallback inside the hasattr(assistant_message, "reasoning_content") block. On fresh tool-call responses where the SDK object didn't expose that attribute at all, the fallback never ran, so the persisted assistant tool-call message was saved without reasoning_content. On the next replay, DeepSeek thinking mode rejected the request with HTTP 400:

The reasoning_content in the thinking mode must be passed back to the API.

What changed:

• move the empty-string fallback to run after tool_calls are materialized into msg
• apply the same creation-time protection to both _needs_deepseek_tool_reasoning() and _needs_kimi_tool_reasoning()
• keep explicit reasoning_content untouched when providers return it

Verification:

• python -m pytest tests/run_agent/test_deepseek_reasoning_content_echo.py -q
• local minimal reproduction before fix: _build_assistant_message(..., tool_calls=...) produced no reasoning_content
• local minimal reproduction after fix: fresh tool-call assistant message includes reasoning_content: ""

[alt-glitch]

Related to #15407 (merged, same reasoning_content echo fix) and #15478 (open, broadens guard to ALL assistant messages). This PR covers the fresh tool-call creation path specifically — may overlap with what #15407 already merged. Please confirm what gap remains vs the merged fix.

[teknium1]

Closing as redundant — the DeepSeek reasoning_content thinking-mode 400 and cross-provider leak chain of issues is now fully covered on main:

• Direct provider detection + same-provider tool-call pad: commits 93a2d6b, d58b305, ad0ac89, 5ae6081
• All-assistant-messages pad rule: commit ad0ac89
• Ordering / cross-provider isolation: commit 9daa062 + regression guard 63bf7a2
• Cross-provider leak (MiniMax reasoning → DeepSeek): PR fix(agent): block cross-provider reasoning leak to DeepSeek/Kimi (#15748) #16500 (merging shortly)

21 regression tests in tests/run_agent/test_deepseek_reasoning_content_echo.py + 2 new tests for the cross-provider scenario exercise every known path. Thanks for the submission — appreciate the digging on this area.