Changing rounding mode of "mul" (*) for Fixed to RoundNearest

[kimikage]

The current rounding mode of "mul"(*) for Fixed is RoundNearestTiesUp. (The comment "rounding up" is ambiguous, though.) In addition, the implementation with RoundDown is commented out. (The comment "truncation" is also ambiguous, though.)

FixedPointNumbers.jl/src/fixed.jl

Lines 76 to 79 in cb29336

	# with truncation:
	#*(x::Fixed{T,f}, y::Fixed{T,f}) = Fixed{T,f}(Base.widemul(x.i,y.i)>>f,0)
	# with rounding up:
	*(x::Fixed{T,f}, y::Fixed{T,f}) where {T,f} = Fixed{T,f}((Base.widemul(x.i,y.i) + (one(widen(T)) << (f-1)))>>f,0)

However, considering the future addition of checked_mul, I think it would be better to change the rounding mode to the default (i.e. RoundNearest) for consistency.

On the other hand, I would also like to have the interfaces to implementations with other rounding modes, for speed. The problem is that there is no official interface that takes a rounding mode as an argument. The third argument of Base.*() collides with z in x * y * z (Of course, there are no arithmetic operations for RoundingMode, though).

[kimikage]

It might be a temporary option to add a function with a redundant name such as mul_with_rounding.