Fix auto-continue bug: don't mark tool calls consumed prematurely

The bug: When the LLM emitted multiple tool calls in one response (e.g.,
str_replace followed by shell), only the first tool was executed. The
remaining tools were lost because mark_tool_calls_consumed() was called
BEFORE processing, marking ALL tools as consumed even when only ONE was
being processed.

This caused has_unexecuted_tool_call() to return false after executing
the first tool, so the parser was reset and the remaining tool calls
were discarded. The auto-continue logic never triggered because it
thought all tools had been handled.

The fix: Remove the premature mark_tool_calls_consumed() call. The
existing logic at line 4696-4699 already handles marking tools as
consumed AFTER execution, and correctly checks for remaining unexecuted
tools before deciding whether to reset the parser.

crates/g3-core/src/lib.rs

@@ -4218,12 +4218,6 @@ impl<W: UiWriter> Agent<W> {
                                 completed_tools.into_iter().take(1).collect()
                             };
 
-                        // Mark tool calls as consumed so has_unexecuted_tool_call() won't
-                        // return true for tools we're about to execute
-                        if !tools_to_process.is_empty() {
-                            parser.mark_tool_calls_consumed();
-                        }
-
                         // Helper function to check if two tool calls are duplicates
                         let are_duplicates = |tc1: &ToolCall, tc2: &ToolCall| -> bool {
                             tc1.tool == tc2.tool && tc1.args == tc2.args