HSI->RGB conversion may return a value out of [0,1]

[kimikage]

FWIW, HSI->RGB conversion may return a value out of [0,1].

julia> convert(RGB, HSV(90,1,1))
RGB{Float32}(0.5f0,1.0f0,0.0f0)

julia> convert(RGB, HSL(90,1,1))
RGB{Float32}(1.0f0,1.0f0,1.0f0)

julia> convert(RGB, HSI(90,1,1))
RGB{Float32}(1.0f0,2.0f0,0.0f0)

This is not consistent with other conversions to RGB. However, the conversions from HSV and HSL also do not check the input range.

julia> convert(RGB, HSV(400,2,2))
RGB{Float32}(2.0f0,0.66666603f0,-2.0f0)

julia> convert(RGB, HSL(400,2,2))
RGB{Float32}(0.0f0,1.3333333f0,4.0f0)

julia> convert(RGB, HSI(400,2,2))
RGB{Float32}(5.26083f0,2.73917f0,-2.0f0)

IMO, it makes sense to clamp the values which are not numerically meaningful or the values ​​which exceed typemin/typemax, but otherwise it is unnecessary.
If we consider color gamuts other than sRGB, we will need to reconsider the clipping.

[kimikage]

One use case where this problem become tangible is calling hex.

julia> hex(HSI(90,1,1))
ERROR: ArgumentError: element type FixedPointNumbers.Normed{UInt8,8} is an 8-bit type representing 256 values from 0.0 to 1.0,
  but the values (1.0f0, 2.0f0, 0.0f0) do not lie within this range.
  See the READMEs for FixedPointNumbers and ColorTypes for more information.

I removed the clamping, i.e. stopped using lerp, in PR #387, because:

• the clamping is done in XYZ/YIQ/YCbCr --> RGB conversion, i.e. Lab, Luv and so on also do not require the extra clamping
• the normal HSV and HSL space do not exceed the normal RGB space
• RGB colors should be used in [0,1], so, the clamping is the responsibility of the user

[kimikage]

I'm going to add the clamping in the HSI-->RGB conversion. (cf. #407)