Tuning nth root for finite fields

[roed]

As of #7931, Sage uses an algorithm due to Johnston for computing the nth root of finite field elements and elements modulo n. In GF(p) for very large p and small n this algorithm is inferior to just factoring x^n-a, since it requires a primitive root modulo p. Preliminary timings indicate that Johnston's algorithm is sometimes faster even in the range of 80 decimal digits, but it sometimes fails spectacularly with runtime 300 times slower than factoring the polynomial.

We should add the polynomial option as an algorithm to nth root and have a reasonable default based on the size of n and p.