fix: document cost estimation review and update default model fallback (#228)

Confirms char-based estimation is unaffected by the CLI v2.1.75 token
overcounting fix. Updates default model fallback to claude-sonnet-4-6.

src-tauri/src/wsl_bridge.rs

@@ -800,14 +800,17 @@ impl WslBridge {
             }
 
             let input_chars = stats.current_request_input.as_ref().map(|s| s.len() as u64).unwrap_or(0);
-            let model = stats.model.clone().unwrap_or_else(|| "claude-sonnet-4-5-20250929".to_string());
+            let model = stats.model.clone().unwrap_or_else(|| "claude-sonnet-4-6".to_string());
 
             (input_chars, stats.current_request_output_chars, stats.current_request_thinking_chars, stats.current_request_tools.clone(), model)
         };
 
         tracing::info!("[COST ESTIMATION] Estimating cost for interrupted request");
 
-        // Use conservative 3.5 chars/token for estimation (vs standard 4)
+        // Char-based estimation: 3.5 chars/token is intentionally conservative (standard English
+        // averages ~4 chars/token). CLI v2.1.75 fixed over-counting in Claude Code's compaction
+        // logic but does not affect this heuristic — we count characters ourselves, independently
+        // of Claude Code's internal token tracking. The 1.2 safety margin avoids undercharging.
         let estimated_input_tokens = (input_chars as f64 / 3.5).ceil() as u64;
         let estimated_output_tokens = ((output_chars as f64 / 3.5).ceil() as u64)
             + ((thinking_chars as f64 / 3.5).ceil() as u64);
@@ -1585,7 +1588,7 @@ fn process_json_line(
                     stats_guard.model.clone()
                 }).unwrap_or_else(|| {
                     tracing::warn!("No model info available for cost calculation, using default");
-                    "claude-sonnet-4-5-20250929".to_string()
+                    "claude-sonnet-4-6".to_string()
                 });
 
                 // Calculate cost for historical tracking (including cache tokens)