Document and warn when solve_poly_system misses solutions not expressible by radicals

[symbolique]

Currently there is no documentation at all of the fact that solve_poly_system will throw away solutions not expressible by radicals. This results in confusion: for example bug #11179 shows some people are unaware.

Furthermore it would be a good idea to emit a warning in that case: roots() could accept an optional argument such as warn=True. Maybe it should even be the default when calling roots()?

[smichr]

cf #12345