Solidify blobs dtype inference

[aarchiba]

Reduce the chance of user surprise due to blob dtype inference - strings are object dtypes, object dtypes don't lead to backend weirdness, ragged blob dtypes raise numpy.VisibleDeprecationWarning unless blobs_dtype is specified. Also fixes problems where grow used to re-infer blob dtypes.

Closes #361

[dfm]

This looks great!

Like you suggested in #361, perhaps we should catch the cases where this deprecation hits and a user hasn't provided a blobs_dtype and provide a recommendation? Happy to merge this and open a separate issue if you think that's best.

Thanks!!

[aarchiba]

https://numpy.org/neps/nep-0034-infer-dtype-is-object.html specifies the deprecation process, so the same change should allow transforming the ValueError into a more informative one as well, for users with a whiz-bang future version of numpy that does that.

[aarchiba]

Ready for review.

[dfm]

Sorry I've been slow to respond - this week has been crazy! I'll get back here in the next few days.

[dfm]

Sorry about the delay! I think this looks great and I'm happy to merge. Thanks for this! I'll start a new pull request thread to draft a new release.