Consider changing output type of ArgMax/Argmin to int32, or allow passing output_type

[philloooo]

Currently WebNN specifies ArgMax/Min returns int64.

• CoreML supported return type is int32 or uint16.
• tflite supports int32, int64.
• dml supports int32, int64, uint32, uint64.

Returning int64 can't be emulated on CoreML:

• the cast operator only supports casting to fp16, fp32, i32, bool.
• None of the operations support int64 as input(so the output of argMax/argMin can't be connected to the next layer if it's int64)
• Graph output doesn't support i64. Only fp16, fp32, i32 is supported.

Given int32 is the intersection across these backends, I see two options to make it work:
a. Update argMin/Max to always output int32
b. On the spec level, allow passing a param of output_type. And allow probing the supported data types for this param.

[philloooo]

Following up from the working group meeting, I am happy to explore option 2 if @fdwr can provide examples on when int64 is useful.

I didn't fully understand your gather example, because I actually don't understand why indices allow both int64 and uint32. The indices should point to valid indices that's within MLOperand's dimensions right? And the dimensions are uint32...