Default to equicoordinate quantiles for multivariate distributions

[mitchelloharawild]

The quantile() function is to return a single point for each dimension of the distribution (as opposed to e.g. hdr()).
For univariate distributions, "equicoordinate" and "marginal" quantiles are semantically and statistically identical. For multivariate distributions, I believe it's more appropriate to default to "equicoordinate" quantiles.

This breaking change introduces reverse dependency check failures with {ggdist} since it assumes marginal quantiles from multivariate distributions are provided. I attempted to make this change for v0.5.0, but reverted it to not break packages (10e8ef9) and am once again reverting it for a v0.6.0 release 😅 (8db6d87).

This reversion is because I believe the API for quantile types could also be improved. The type argument currently used in dist_multivariate_normal() causes semantic conflict with the type argument of quantile.default() which is an integer 1-9 for the algorithm to use. As such I've renamed this usage as kind, which ggdist should probably use quantile(kind = "marginal") when working with distributional objects since its geometries are univariate/marginal in nature (@mjskay - the v0.7.0 release will change the default for multivariate distributions to equicoordinate).