together, cancel, and a zero test for rational functions #2146

Most proofs in Modenov's Problems in Geometry (#2131) end in "this rational function of a, b, c is 0". Readers ran more than 200 of them through Simplify. The failures they checked at exact random points were all 0, so the identities are true and the failures are the library's:

• fractions over one denominator are not combined (Simplify does not combine fractions over a common denominator #2137);
• the answer depends on the order or the names of the variables;
• Simplify is not idempotent (Simplify can stop before its own answer is simplified #2133);
• some time out at 20 s.

The parts exist: the polynomial layer, its multivariate GCD (MathS.Polynomials.Gcd), and Transformation.RationalCanonicalization, which proves identities Simplify does not.

Wanted, for 2.8, as explicit operations:

• together(expr): one numerator over one denominator;
• cancel(expr): that, with the GCD divided out;
• a zero test built on them.

Whether Simplify should call them by default is a cost and canonical-form question, so it belongs with v3. The trigonometric identities that fail (#2135, #2142) would follow from the same zero test, reduced modulo sin² + cos² − 1.

Related: #1251.