Reject out-of-range num_samples in the op input parsers

[mhucka]

This fixes b/510431411, "Uninitialized Tensor Heap Memory Leak via Integer Overflow in BalanceTrajectory", which is an instance of CWE-200/CWE-908. The problem is this: a num_samples value near INT_MAX overflows the int arithmetic (num_samples + num_threads - 1)/num_threads in BalanceTrajectory and in the noisy ops' ComputeSmall. The resulting negative chunk sizes make every worker thread skip its sampling loop, so TfqNoisySamples returns its output tensor unwritten.

This PR adds a check on number of samples in GetNumSamples and GetIndividualSample. The latter previously had no bound at all; a negative value reached TensorShape::AddDim, which would CHECK-fail and terminate the process. It now returns InvalidArgument.

The real changes are in the files tensorflow_quantum/core/ops/parse_context.cc and tensorflow_quantum/core/ops/parse_context.h. The remaining file changes are to test files.

[gemini-code-assist]

Code Review

This pull request introduces an upper bound of 10,000,000 (kMaxNumSamples) on the number of samples accepted for any single circuit or Pauli sum to prevent integer overflow during multi-threaded work splitting. It updates the validation logic in GetNumSamples and GetIndividualSample to enforce this limit and adds corresponding unit tests across several simulation and noisy operator test suites. No review comments were provided, so there is no additional feedback to address.