Separate methods for ^ to use power_by_squaring or pow!

[lbenet]

No description provided.

[lbenet]

This implements a suggestion of @dpsanders (off-line discussion)

[lbenet]

This makes that (interval(-1,1) + Taylor1(5)^4 yields the expected result.

[coveralls]

Coverage decreased (-0.2%) to 98.085% when pulling 8141256 on lb/power_by_squaring into 81b4649 on master.

[dpsanders]

Maybe add some tests with intervals? That would require adding IntervalArithmetic.jl to test/REQUIRE.

[lbenet]

Good point; I'll implement some tests for intervals.

[lbenet]

Due to the use of power_by_squaring, some intervals arising from (t-a)^n (a is an Interval) are actually wider than the explicit formula, which evaluates a^k with appropriate k. This is simply the dependency problem.

[lbenet]

I think this is ready for review.

[lbenet]

Tests are passing and 4d5ac34 provided narrower intervals.