Proposal update change

[segasai]

This is a change trying to address how tuning of proposal distributions is done. Previously only 'blob' returned from one every 'queue_size' was used to adjust the tuning params. With this change the update_PROPOSAL() are called every time to mostly accumulate statistics on number of acceptances and rejects and the actual update will happen still every queue_size.

The advantages are

• the scale adjustment can become much less noisy. I.e. if some parts of the posterior may require smaller scale. Currently that info may be missed, because only one of every queue_size results is analyzed
• We can react by switching to doubling scheme for slice sampling even if just a single likelihood evaluation had an issue

Disadvantages:

• change in the update_() interface
• a little bit more CPU overhead
• additional attributes in the sampler

I think the advantages are worth it, but I think some refactoring may be warranted, like grouping the samplers in a separate classes, so they can keep the information there

[coveralls]

Pull Request Test Coverage Report for Build 2904104172

• 39 of 41 (95.12%) changed or added relevant lines in 2 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+0.04%) to 90.53%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
py/dynesty/nestedsamplers.py	36	38	94.74%
<!--	Total:	39	41	95.12%	-->

Totals
Change from base Build 2884196862:	0.04%
Covered Lines:	3738
Relevant Lines:	4129

---

💛 - Coveralls