Fixes #16478

Problem

All tool result messages (role: "tool") were missing the required name field. While OpenAI's API tolerates this, Gemini and other strict OpenAI-compatible providers return HTTP 400 when name is absent.

Root Cause

The tool message dicts in run_agent.py only included role, content, and tool_call_id, but omitted the name field that identifies which function the tool response belongs to.

Fix

Added "name": <function_name> to all 9 tool message construction sites in run_agent.py:

1. Context sanitizer stub (line ~4716) — pre-call gap-filler
2. Argument repair insertion (line ~8248) — corrupted JSON repair
3. Parallel pre-flight interrupt (line ~8568) — user interrupt during parallel
4. Parallel execution result (line ~8834) — main parallel tool result
5. Sequential interrupt mid-tool (line ~8872) — user interrupt during sequential
6. Sequential execution result (line ~9199) — main sequential tool result
7. Sequential interrupt after-tool (line ~9227) — user interrupt after partial completion
8. Invalid tool name error (line ~12059) — unknown tool self-correction
9. Invalid JSON args error (line ~12151) — malformed arguments self-correction

Testing

• All 10 role: "tool" messages now include the name field
• Syntax verified with ast.parse() on the full file
• No functional changes to tool execution logic — only message formatting