[Frontend] Update SProd observable tracing (#935)

**Context:** After change 296316654bd6aabb3ff67eb0bac440fa8d706ee8 in
PennyLane, the SProd operation will return the flattened terms.

**Description of the Change:** This change fixes the tracing of SProd to
trace over all terms returned by `obs.terms()` as opposed to only the
first term.

**Benefits:** Fixes latest/latest/latest

**Possible Drawbacks:** None

**Related GitHub Issues:** 

[sc-69025]

.dep-versions

@@ -5,3 +5,4 @@ llvm=cd9a641613eddf25d4b25eaa96b2c393d401d42c
 enzyme=v0.0.130
 
 # Always remove custom PL/LQ versions before release.
+pennylane=296316654bd6aabb3ff67eb0bac440fa8d706ee8

doc/changelog.md

@@ -63,7 +63,14 @@
 
 * Using float32 in callback functions would not crash in compilation phase anymore,
   but rather raise the appropriate type exception to the user.
-  [(#916)]https://github.com/PennyLaneAI/catalyst/pull/916
+  [(#916)](https://github.com/PennyLaneAI/catalyst/pull/916)
+
+* Fix tracing of `SProd` operations
+  [(#935)](https://github.com/PennyLaneAI/catalyst/pull/935)
+
+  After some changes in PennyLane, `Sprod.terms()` returns the terms as leaves
+  instead of a tree. This means that we need to manually trace each term and
+  finally multiply it with the coefficients to create a Hamiltonian.
 
 <h3>Internal changes</h3>
 

frontend/catalyst/jax_tracer.py

@@ -705,10 +705,13 @@ def trace_observables(
         nested_obs = [trace_observables(o, qrp, m_wires)[0] for o in obs]
         obs_tracers = hamiltonian_p.bind(jax.numpy.asarray(jnp.ones(len(obs))), *nested_obs)
     elif isinstance(obs, qml.ops.op_math.SProd):
-        terms = obs.terms()
-        coeffs = jax.numpy.array(terms[0])
-        nested_obs = trace_observables(terms[1][0], qrp, m_wires)[0]
-        obs_tracers = hamiltonian_p.bind(coeffs, nested_obs)
+        coeffs, terms = obs.terms()
+        coeffs = jax.numpy.array(coeffs)
+        nested_obs = []
+        for term in terms:
+            obs = trace_observables(term, qrp, m_wires)[0]
+            nested_obs.append(obs)
+        obs_tracers = hamiltonian_p.bind(coeffs, *nested_obs)
     else:
         raise NotImplementedError(
             f"Observable {obs} (of type {type(obs)}) is not impemented"