Fix out-of-bounds error

[apsonawane]

This pull request adds input validation to the RaggedTensorToSparse operator to ensure the row-splits tensor is well-formed. These checks help catch malformed inputs early and improve error messaging.

Input validation improvements:

• Added a check to ensure the row-splits tensor is not empty, starts at zero, and is monotonic non-decreasing, with clear error messages for each case in op_ragged_tensor.cc.

[Copilot]

Pull request overview

This pull request hardens the RaggedTensorToSparse custom operator by validating the row-splits (nested splits) input before it is used to compute output shapes and index buffers, preventing invalid shapes/indices and catching malformed inputs earlier.

Changes:

• Added a guard for an empty row-splits tensor (prevents p_n_elements[-1] access when the input has 0 elements).
• Added validation that row-splits starts at 0.
• Added validation that row-splits is monotonic non-decreasing.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.