Better name for default unimodal HDI method

[OriolAbril]

See discussion in https://github.com/arviz-devs/arviz-stats/pull/28

[sethaxen]

So I've come to the point where I actually think the original unimodal is best for a few reasons:

• The distributions where it's expected to be a good estimate for all probabilities are precisely the unimodal ones.
• While contiguous better describes the property the interval has, unimodal better describes the assumed property of the distribution.
• Useful by comparison to our multimodal methods.
• Consistent with old ArviZ