Commonize code between Fixed and Normed

[kimikage]

This PR is the first part of #139.
I am very sorry about a conflict with #143, but in the long-term, I think this PR will help enhance the functions (e.g. checked arithmetic).

Because there are many changes, it is recommended that you track changes on a per-commit basis.

This PR includes a minor breaking change, i.e., this change the exception type of one(Q0f7) etc. from InexactError to ArgumentError.

I'm ready to fix isinteger (issue #120, PR #123) after the merging this PR. And then, I'll fix the issue #150.

Edit:
Rescheduling: this -> issue #153 -> isinteger -> rem -> issue #157

[timholy]

This is very nice, I strongly support this.

Are the methods removed in 2e19e67 really "useless" without a2e3323? (I haven't checked myself.)

[timholy]

Suggested change

	significand = xu | (one(xu) << significand_bits(Tf))
	significand = xu | (oneunit(xu) << significand_bits(Tf))

(I failed to notice this earlier.)

[kimikage]

Do we need to worry about the difference between one and oneunit for types without dimensions?
Anyway, I changed one to oneunit for future consistency.

[timholy]

Not so far, but in cases where you mean "1 of the given type" then oneunit is appropriate, no matter whether it differs from one.

one is a little weird because it could pretty much be defined as the constant 1.

[kimikage]

Are the methods removed in 2e19e67 really "useless" without a2e3323?

FixedPointNumbers.jl/src/FixedPointNumbers.jl

Lines 34 to 37 in 7ad0f0c

	# comparison
	==(x::T, y::T) where {T <: FixedPoint} = x.i == y.i
	<(x::T, y::T) where {T <: FixedPoint} = x.i < y.i
	<=(x::T, y::T) where {T <: FixedPoint} = x.i <= y.i

FixedPointNumbers.jl/src/normed.jl

Lines 254 to 256 in 7ad0f0c

	# Comparisons
	<(x::T, y::T) where {T <: Normed} = reinterpret(x) < reinterpret(y)
	<=(x::T, y::T) where {T <: Normed} = reinterpret(x) <= reinterpret(y)

They don't seem to differ in essence.
However, the tests of FixedPointNumbers.jl are somewhat poor, so we should to recheck the codes before release.

[kimikage]

I added tests for reinterpret, although they have nothing to do with my changes directly.

Edit:
AppVeyor seems to be stuck.:sweat_smile:
AppVeyor seems to be stuck again on master branch.:cold_sweat:

I have confirmed that this has passed in my forked repository.
kimikage@e18c350
https://ci.appveyor.com/project/kimikage/fixedpointnumbers-jl/builds/29815705