Replace using_pjrt() xla runtime device_type() check with in xla.py

Fixes Lightning-AI#20419

`torch_xla.runtime.using_pjrt()` is removed in pytorch/xla#7787

This PR replaces references to that function with a check to [`device_type()`](https://github.com/pytorch/xla/blob/master/torch_xla/runtime.py#L83) to recreate the behavior of that function, minus the manual initialization

Author: byi8220

src/lightning/fabric/accelerators/xla.py

@@ -109,7 +109,7 @@ def _using_pjrt() -> bool:
     if _XLA_GREATER_EQUAL_2_1:
         from torch_xla import runtime as xr
 
-        return xr.using_pjrt()
+        return xr.device_type() is not None
     from torch_xla.experimental import pjrt
 
     return pjrt.using_pjrt()