[Spec issue] Rounding or truncation behavior for Cast operator is not specified in the spec

[max-ku]

• ORT performs rounding when Cast operator is used from f32 to bfloat16.
• There is no explicit mentioning about rounding or truncation behavior for Cast operator on documentation page.
• The test example on same page related to Cast operator performs truncation, not rounding.

Suggesting to revise documentation:

• explicitly state that rounding is performed in Cast operator for floating-point conversions.
• update example to remove confusion about truncation behavior for Cast from f32 to bfloat16.

Repro for rounding
rounding.zip

[jcwchen]

Hi @max-ku,
I also think the document should be clearer for this case... There are a few similar issues related to Cast like #2285. I will add this item into tracking. Feel free to propose a PR for improving document if you have bandwidth. Thank you!