Major changes to `mask.bubbles_mask_nonzero()`:
* Now supports non-integer products of `sigma` and `max_sigma_from_nonzero`. The previous version assumed that `sigma * max_sigma_from_nonzero` would always produce an integer.
* Bubble locations are now sampled from pixels identified via binary dilation, with random noise between `-0.5` and `0.5` around the centre of the pixel.
* `max_sigma_from_nonzero` can now be an array of values, of the same length as `sigma`, for bubble-specific constraints.
* `max_sigma_from_nonzero` can be `0` or `np.inf`.
Minor changes:
* `mask.bubbles_mask_nonzero()` now passes `**kwargs` to `mask.bubbles_mask()` and `build.build_mask()`
* `mask.bubbles_mask()` now passes `**kwargs` to `build.build_mask()`
* `build.build_mask()` now has defaults of `scale=True`and `sum_merge=False`. These defaults were originally set in `mask.bubbles_mask()` and `mask.bubbles_mask_nonzero()`.
* Added timing demo to `examples.py`, comparing `bubblemask`'s density method to a convolution method.
* Improved all documentation, including using `numpy` style
* Arguments `scale` and `sum_merge`, which should now be passed as `**kwargs`, have been removed from documentation for `mask.bubbles_mask_nonzero()`.