Fix inconsistencies with Time{WithZone}#{hash,eql?}

[marcandre]

If foo.eql?(bar) returns true then so should bar.eql?(foo).
Moreover, foo.hash must then be the same as bar.hash.
Otherwise the Hash lookup will fail. TimeWithZone wants to act like a Time, but currently:

bar = Time.utc(2000)
foo = bar.in_time_zone

foo.eql?(bar) # => true
bar.eql?(foo) # => false
foo.hash == bar.hash # => false (only in Ruby 1.9)

This pull request fixes this. Since TimeWithZone wants to be freely replaceable with a Time:

bar.eql?(foo) # => now true
foo.hash == bar.hash # => now true (even in Ruby 1.9)

There is also a problem with DateTime vs TimeWithZone:

dt = foo.to_datetime
foo.eql?(dt)  # => true
dt.eql?(foo)  # => false
dt.hash == foo # => false.

Since DateTime and a Time are both builtin classes and that a Time is never eql? to a DateTime or vice-versa (as per the doc and the implementation), the patch also insures that a TimeWithZone can no longer be eql? to a DateTime:

foo.eql?(dt)  # => now false

Thanks

Extended discussion at #905