Fix dtype comparison using 'is' instead of '==' in _integrate.py

- Replace 'is' with '==' for dtype comparisons on lines 265 and 388
- Fixes assertion errors in scenarios involving serialization/distributed computing
- Ensures robust dtype comparison regardless of how dtype objects are created
- Maintains backward compatibility and intended behavior
Fixes patrick-kidger#678

Author: ArianAmani

diffrax/_integrate.py

@@ -262,7 +262,7 @@ def _clip_to_end(tprev, tnext, t1, keep_step):
     # The tolerance means that we don't end up with too-small intervals for
     # dense output, which then gives numerically unstable answers due to floating
     # point errors.
-    if tnext.dtype is jnp.dtype("float64"):
+    if tnext.dtype == jnp.dtype("float64"):
         tol = 1e-10
     else:
         tol = 1e-6
@@ -385,7 +385,7 @@ def body_fun_aux(state):
             error_order,
             state.controller_state,
         )
-        assert jnp.result_type(keep_step) is jnp.dtype(bool)
+        assert jnp.result_type(keep_step) == jnp.dtype(bool)
 
         #
         # Do some book-keeping.