Define Dimension formatters for numpy integer types

[philippjfr]

As described in #975, widgets do not correctly format integers at the moment. This is easily fixed by defining Dimension.type_formatters for the numpy integer types.

[jlstevens]

Looks good and I a happy to merge once the tests are passing.

The only suggestion I have is to investigate whether there is an easy way to iterate over all numpy integer types - if so, iterating over them would be better than having to list them all explicitly this way.

[philippjfr]

The only suggestion I have is to investigate whether there is an easy way to iterate over all numpy integer types.

As long as we are treating them all as separate types I prefer to be explicit about it. I do think that rather than specifying it by an actual type it might be more useful to declare that all integer types should respect the formatter. May we should consider whether int and float formatters should also apply to all numpy types where dtype.kind=='i' and dtype.kind=='f' respectively, if no more specific formatter has been set.

[philippjfr]

Apparently this has resulted in small changes in the bokeh table tests. Will update.

[philippjfr]

@jlstevens Any opinions on my comment above?

[jlstevens]

...it might be more useful to declare that all integer types should respect the formatter

I think I agree that this makes sense.

May we should consider whether int and float formatters should also apply to all numpy types where dtype.kind=='i' and dtype.kind=='f' respectively, if no more specific formatter has been set.

Sounds reasonable for scalars...I assume the formatters won't do a good job for arbitrary arrays.

Edit: Note that I am happy to merge if you want to tackle these other ideas later.

[philippjfr]

Let's go with this for now. Rebuilding test data shortly.

[jlstevens]

The pr build passed. Merging.