Add roundEven operator

[fdwr]

• Fixes Rounding operators #817.
• Addresses incomplete quantizeLinear emulation step.

partial interface MLGraphBuilder
{
    ...
+   MLOperand roundEven(MLOperand input, optional MLOperatorOptions options = {});
    ...
};

Mappings to backend functions:

RHE round halves to nearest even:

• DML ROUND with DML_ROUNDING_MODE_HALVES_TO_NEAREST_EVEN
• tfl.round (TFL::RoundOp) 🤔 *It's not clearly documented for TFLite, but I'm presuming TFL follows TF in using RHE.
• ONNX Round
• PyTorch torch.round
• ❔coremltools.converters.mil.mil.ops.defs.iOS15.elementwise_unary.round 🤔 *From Ningxin's investigation, CoreML's rounding mode behaves inconsistently depending on whether it goes through the NPU vs GPU, but it should be consistently emulatable via elementwise_binary.sub(x, elementwise_binary.mod(x, 1.0f)). 🤞

---

Preview | Diff

[huningxin]

LGTM!

[fdwr]

LGTM!

👍 I'll give it a few days for others to look at too lest they have feedback. 🔍⏳

FYI @reillyeon @philloooo.