[CK_TILE] Implement fp8 quant tests/examples for layernorm and rmsnorm

[ruanjm]

• Compile option --offload-compress is added because the code object is too large. I'm not sure whether this flag would affect performance.
• Y scale base of fp8 is set as 240 which is different from that of int8 (127). This base value kMaxY is hard coded in host code because static_cast in host side cannot interpret fp8 to float correctly.
• Outputs of check_err() for fp8 is improved.