Let's change the default for n_initial_points to 5

[SRFU-NN]

It is 10, which is way too much for simple spaces. This might also help with the changed behaviour in ask-and-tell-ipynb #200

[dk-teknologisk-mon]

I think that in the ideal world the number should be dynamic and depend on the size of the problem. If the user carries out optimization blindly, they shouldn't really care when they transition from initial points to the use of an acquisition function. If they know what they are doing, then they can change behavior by providing a number themselves. In the Brownie Bee user interface, we use the rule max(5, n_dimensions+1), though simulations on benchmarks suggest that n_dimensions*3 is on average a more efficient strategy, even in relatively easy systems.

In practice, new users are often more scared of a large number of being shown a large number of experiments up front, which is why Brownie Bee does what it does. If a user asks for experiments one at a time without any thought, then the choice here doesn't matter much.