fix: model_token_limit catch-all returns None for unknown models

Restores the fix from @Offwhite-Del that was lost during rebase.
Unknown models now return None so preflight skips the context-window
check and lets the API enforce its own limits.

rust/crates/api/src/providers/mod.rs

@@ -689,12 +689,9 @@ pub fn model_token_limit(model: &str) -> Option<ModelTokenLimit> {
             max_output_tokens: 16_384,
             context_window_tokens: 200_000,
         }),
-        // Generic fallback for any model: assume 128K context, 8K output.
-        // This prevents the "unknown model → no limit check → context overflow" bug.
-        _ => Some(ModelTokenLimit {
-            max_output_tokens: 8_192,
-            context_window_tokens: 131_072,
-        }),
+        // Unknown models return None so preflight skips the context-window
+        // check and lets the API enforce its own limits.
+        _ => None,
     }
 }