[doc] Add a note about the offline cache in the print_ir part (taichi…

…-dev#5986)

* Add a note about the offline cache 

Add a note about the offline cache in the `print_ir` part to help users.

* Update global_settings.md

docs/lang/articles/reference/global_settings.md

@@ -26,7 +26,7 @@ In case you want to use taichi cuda backend together with GGUI on a machine with
   errors: `ti.init(advanced_optimization=False)`.
 - Disable fast math to prevent possible undefined math behavior:
   `ti.init(fast_math=False)`.
-- To print intermediate IR generated: `ti.init(print_ir=True)`.
+- To print intermediate IR generated: `ti.init(print_ir=True)`. Note that compiled kernels are [cached by default](https://docs.taichi-lang.org/docs/performance#offline-cache). To force compilation and IR emission, use `ti.init(print_ir=True, offline_cache=False)`.
 
 ## Runtime