Validate tool execution results to prevent silent failures

[ssevera1]

What

Add validation that tool execution results are not None before adding to conversation.

Why

Prevents silent failures when tools crash without returning a ToolResult object.

[claude]

Checked the guard against the surrounding control flow in agent/core/engine.py: it sits after the try/except/finally so the spinner is always stopped, mirrors the existing "Unknown tool" early-return pattern, and both call sites (lines 116 and 128) ignore the return value, so the early return changes nothing else. Without it, a tool that returns None instead of a ToolResult would hit format_tool_result's result.success and raise an AttributeError outside the try block, or record the literal string "None" into the conversation — the guard turns that into a recorded error message instead. All execute implementations are annotated -> ToolResult, so this only fires on a genuine tool bug; scoped to 7 lines in one file and matches the PR description.