Deprecate convertFp16() in favor of convertTo() with CV_16F #24909
Description
On 4.x branch we have CV_16F support, so we don't need to emulate FP16 buffers as CV_16S.
Tasks:
- deprecate convertFp16()
- improve OpenCL performance of convertTo()
- replace usage of convertFp16() in DNN module
- replace usage of convertFp16() in other modules (imgcodecs?)
- deprecate CUDA's convertFp16()
- remove convertFp16() in 5.x
P.S. convertFp16() API returns CV_16S buffer by default.