[BugFix][Opencl] Explicitly cast min/max operands

[hope51607]

I encountered the same max / min ambiguous issue caused by the ternary operator as #2821 and #5090.

The reason is that the type of the ternary operator is not unchar.

The patch that fixes this problem seems to be missing in #5107 .
This pr simply adds them back.

[shingjan]

Thanks for the work! I happened to notice that the test introduced in #5090 is disabled now. Can you confirm that this test passes with this PR? @kazum It would be great if you could take a look as well!

cc: @tqchen @junrushao1994

[masahi]

@hope51607 Can you also add fp16 tests? I've met the "type ambiguity" error when trying to compile fp16 models for opencl before. I assume this patch fixes such issue?

[hope51607]

@shingjan I have enabled these tests and confirmed that they passed.
@masahi I tried to add fp16 test, but in my environment it does not seem to support half-precision floating point.
I am confused whether to add it.

[masahi]

Sure, no problem.