NNETAR doc improvements

[robjhyndman]

1. The times argument in fable::forecast.NNETAR() specifies default 1000, but it is actually 5000, as it is passed through from fabletools::forecast.mdl_ts().
2. Better explanations needed about the difference between simulate=TRUE and bootstrap=TRUE. simulate draws samples from a normal distribution, whereas boostrap resamples from the residuals. Both result in simulated future sample paths from which empirical forecast distributions are constructed.
3. Setting simulate=FALSE is ignored.
4. The times argument description suggests simulated intervals are optional.
5. Perhaps explain that setting times=0 will lead to degenerate distributions equal to the mean.

[mitchelloharawild]

1. The `times` argument in `fable::forecast.NNETAR()` specifies default 1000, but it is actually 5000, as it is passed through from `fabletools::forecast.mdl_ts()`.

That's unfortunate... I guess there's not a nice way for method specific defaults unless the generic has times=NULL?

2. Better explanations needed about the difference between `simulate=TRUE` and `bootstrap=TRUE`. `simulate` draws samples from a normal distribution, whereas `boostrap` resamples from the residuals. Both result in simulated future sample paths from which empirical forecast distributions are constructed.

Done.

3. Setting `simulate=FALSE` is ignored.

It's not ignored, it sets times=0

4. The `times` argument description suggests simulated intervals are optional.

They somewhat are optional - if simulate=FALSE then intervals will not be produced.

5. Perhaps explain that setting `times=0` will lead to degenerate distributions equal to the mean.

Sure.

---

Updated in b336612