Handle zero-cycle GPU divisor sessions with heuristics

PerfectNumbers.Core.Tests/MersenneNumberDivisorGpuTesterTests.cs

@@ -142,6 +142,22 @@ public class MersenneNumberDivisorGpuTesterTests
         divisorsExhausted.Should().BeTrue();
     }
 
+    [Fact]
+    [Trait("Category", "Fast")]
+    public void ByDivisor_session_computes_cycle_when_zero_is_provided()
+    {
+        var tester = new MersenneNumberDivisorByDivisorGpuTester();
+        tester.ConfigureFromMaxPrime(43UL);
+
+        using var session = tester.CreateDivisorSession();
+        ulong[] primes = { 31UL, 37UL, 41UL };
+        byte[] hits = new byte[primes.Length];
+
+        session.CheckDivisor(223UL, MontgomeryDivisorData.FromModulus(223UL), 0UL, primes, hits);
+
+        hits.Should().ContainInOrder(new byte[] { 0, 1, 0 });
+    }
+
     [Fact]
     [Trait("Category", "Fast")]
     public void ByDivisor_session_checks_divisors_across_primes()

PerfectNumbers.Core/Gpu/MersenneNumberDivisorByDivisorGpuTester.cs

@@ -471,15 +471,26 @@ public sealed class MersenneNumberDivisorByDivisorGpuTester : IMersenneNumberDiv
             }
 
             ulong cycle = divisorCycle;
-            // EvenPerfectBitScanner resolves divisor cycles before scheduling GPU work, so the zero-cycle fallback stays disabled.
-            // if (cycle == 0UL)
-            // {
-            //     throw new InvalidOperationException("GPU divisor cycle length must be non-zero.");
-            // }
-
             ulong firstPrime = primes[0];
+
+            if (cycle == 0UL)
+            {
+                if (!MersenneDivisorCycles.TryCalculateCycleLengthForExponent(divisor, firstPrime, cachedData, out ulong computedCycle, out bool primeOrderFailed) || computedCycle == 0UL)
+                {
+                    cycle = MersenneDivisorCycles.CalculateCycleLength(divisor, cachedData, skipPrimeOrderHeuristic: primeOrderFailed);
+                }
+                else
+                {
+                    cycle = computedCycle;
+                }
+
+                if (cycle == 0UL)
+                {
+                    throw new InvalidOperationException("GPU divisor cycle length must be non-zero.");
+                }
+            }
+
             // Precompute the initial cycle remainder on the host to keep the GPU kernel branch-free.
-            // EvenPerfectBitScanner resolves divisor cycles before scheduling GPU work, so cycle never equals zero here.
             ulong firstRemainder = firstPrime % cycle;
 
             Monitor.Enter(_lease.ExecutionLock);