com.microsoft.RotaryEmbedding: validate position_ids are within cos/sin cache bounds

[Copilot]

Description

Adds explicit bounds-checking for position_ids in the CPU RotaryEmbedding kernel before indexing into the cos/sin cache. Previously, out-of-range values caused silent UB (out-of-bounds memory reads).

RunRotaryEmbedding (rotary_embedding.cc)

• Format 0 (scalar): rejects if base_pos < 0 or base_pos + sequence_length > max_sequence_length
• Format 1 (2D [batch, seq]): iterates all elements and rejects if any value is outside [0, max_sequence_length)
• Returns INVALID_ARGUMENT with the offending value and its flat index

New CPU tests (rotary_embedding_op_test.cc)

• Format 1: value exceeds max_sequence_length, negative value, OOB value in a batch
• Format 0: base offset + sequence length exceeds cache size, negative base offset

Motivation and Context

When a caller passes position_ids with values outside the cos/sin cache range, the kernel would silently read garbage memory. The op should fail fast with a descriptive error rather than producing corrupted output or crashing.

Note: the CUDA kernel does not yet have equivalent device-side validation and is a follow-up.