SplitToSequence cannot support float16 as input/output

Describe the issue

SplitToSequence operator cannot support float16 as input, although it was said 'yes' in ONNX doc.
It impact fp16 converter. Because the output is a sequence, so user have to convert the element in the output sequence to fp16 one by one, the performance is bad.

To reproduce

1. prepare a [5,5] input array with dtype = np.float16
2. call op.SplitToSequence(input, dim=0, num_outputs=2)
3. the expected output is sequence([2,5], [3,5]).

It works on float32.

Urgency

No response

Platform

Windows

OS Version

11

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

1.14

ONNX Runtime API

Python

Architecture

X64

Execution Provider

Default CPU

Execution Provider Library Version

No response