[Core] Add nesting-depth limit to Loop::validate_and_infer_types and test TI nesting assert (CWE-674)

[axinging]

Details:

Loop::validate_and_infer_types (loop.cpp:206) calls m_bodies[0]->validate_nodes_and_infer_types() which walks every body node and invokes each node's validate_and_infer_types(). If a body node is itself a v5::Loop, validation re-enters the same descent with no depth counter, producing one native C++ stack frame per nesting level. A sufficiently deep chain overflows the call stack and crashes the process (SIGSEGV / EXCEPTION_STACK_OVERFLOW).

This is the same defect shape as the If recursion，at a distinct location (Loop).

TensorIterator already has a hard assert in
revalidate_and_infer_types_for_body_ops() (tensor_iterator.cpp:47) that unconditionally blocks any TI nesting:

OPENVINO_ASSERT(... == nullptr, "No nested TensorIterator");

This is stronger than a depth guard — it forbids TI-in-TI entirely. No code change is needed for TI, but this assert was previously untested.

Fix: use the shared OV_VALIDATION_DEPTH_GUARD macro (introduced in the companion If PR) in Loop::validate_and_infer_types to guard against nesting deeper than kMaxValidationDepth = 64:

OV_VALIDATION_DEPTH_GUARD(this, "Loop");

Tests added:

• type_prop.loop_nested_depth_limit_is_rejected (ov_core_unit_tests) — validates the Loop depth guard throws NodeValidationFailure at depth 1024.
• type_prop.tensor_iterator_nested_is_rejected (ov_core_unit_tests) — locks TI's existing hard assert: nesting TI-in-TI at depth 2 throws AssertFailure with "No nested TensorIterator".

Tickets:

• CVS-192817

AI Assistance:

• AI assistance used: yes
• Bug root cause was found by AI.