comment and code discrepancy

In FPU.scala:
```
// we don't currently support round-max-magnitude (rm=4)
io.illegal_rm := io.inst(14,12).isOneOf(5.U, 6.U) || io.inst(14,12) === 7.U && io.fcsr_rm >= 5.U

```
It seems that rm=4(round to nearesr, with tie to Max magnitude) is currently supported.

Start in FPU.scala by checking the illegal_rm logic against the RISC-V rounding-mode definitions, especially rm=4. Update the discrepancy so the comment accurately describes the behavior, then run the relevant FPU or Rocket Chip tests if available to confirm the change.

Written by the indexing model from the issue text.