[KnownFPClass] Incorrect subnormal handling for `maximum(num)`/`minimum(num)`, and `maxnum`/`minnum`

Discovered by https://github.com/llvm/llvm-project/pull/219091

Under FTPZ/DAPZ, a negative subnormal value could become positive zero.

Currently, the MinMax family of functions will incorrectly rule out positive zero even if negative subnormal outputs/inputs can flush to positive zero.

Start by reviewing the related change in llvm/llvm-project pull request 219091 and trace the KnownFPClass handling for the MinMax family. Reproduce or inspect behavior under FTPZ/DAPZ for negative subnormal inputs and outputs, then verify that maximum(num), minimum(num), maxnum, and minnum handle values that flush to positive zero correctly.

Written by the indexing model from the issue text.