Support propagation of uncertainties in math functions

[nvaytet]

Functions include trigonometric functions, exp, log etc.

[jl-wynen]

We already support exp and log:

scipp/lib/core/include/scipp/core/value_and_variance.h

Lines 90 to 99 in 11d5a99

	template <typename T> constexpr auto exp(const ValueAndVariance<T> a) noexcept {
	using std::exp;
	const auto val = exp(a.value);
	return ValueAndVariance(val, val * val * a.variance);
	}
	template <typename T> constexpr auto log(const ValueAndVariance<T> a) noexcept {
	using std::log;
	return ValueAndVariance(log(a.value), a.variance / (a.value * a.value));
	}

I think we are mainly lacking trigonometric functions. The problem with those is that the propagation needs to know whether the input is in radians or degrees.

[jokasimr]

@damskii9992 In case you need a workaround, you can maybe use something like:

def uncertain_sin(x):
    x = x.to(unit='rad')
    y = sc.sin(sc.values(x))
    y.variances = x.variances * sc.cos(sc.values(x)).values**2
    return y