Support comparison between all Period subtypes

[omus]

Allows user defined Period subtype to be compared against other Period subtypes.

See TimeZones.jl comment

[tkelman]

thanks. could use non-methoderror tests of something this allows that doesn't work yet on master

[JeffBezanson]

Maybe it would work better to have TimePeriod and CalendarPeriod as abstract subtypes of Period, and use TimePeriod to mean what FixedPeriod means now? Whether a period corresponds to a fixed amount of time seems like a pretty important distinction. Then we wouldn't need methods that throw method errors.

[omus]

FixedPeriod also contains Week and Day where as TimePeriod does not. Sometimes it isn't so clear when periods are fixed. In Base.Dates a Day is treated as a period of 24 hours however in TimeZones.jl a Day could mean 23, 24, or 25 hours depending on context.

[JeffBezanson]

The MethodError should contain the arguments, not their types (when possible).

[JeffBezanson]

Ah, I see. But by allowing Days == Seconds aren't we already implicitly declaring Day to be a fixed time period?

[omus]

Currently Base does always treats a Day(1) == Hour(24). My example was just to show how periods people typically consider fixed may not be. Personally I'd like revise period comparison but that's out of the scope of this PR.

[JeffBezanson]

Needs a rebase, then I guess we should merge this.

[omus]

Rebased and squashed commits.