Detect float point type earlier in BGV pipeline

Currently the detection of float point type and the warning for it is in `--secret-to-bgv`. However, earlier pass like `--validate-noise` assumes input in BGV style, and float point input to it will segfault.

As we put the scheme flag in the pipeline like `--mlir-to-bgv`, we know the scheme parameter from the begining so we could reject it earlier.