The rounding mode should be determined by reading the FPU's current
rounding mode.
1. Two different systems for handling rounding is confusing and
excessively complicated.
2. The FPU can be used for handling the lower precision calculations,
where it will use its rounding mode anyway.
3. If specialized template versions of BigFloat will be the native
types, then they'll use the FPU rounding mode.
4. No new API is necessary to get/set the modes.
5. It'll respond appropriately when interfacing with code from other
languages that manipulate the FPU rounding modes.
This goes ditto for the floating point exceptions.
