fix the resuming of dynamic nested sampler runs with the pool

[segasai]

Addresses #438

What remains to be done here is some kind of test that would prevent this bug from happening. I.e. the test that would verify that after restoring with the pool, it's actually used.

If I won't have an idea for the test, I'll be merging without.

[coveralls]

Pull Request Test Coverage Report for Build 4934149245

• 22 of 23 (95.65%) changed or added relevant lines in 3 files are covered.
• 1 unchanged line in 1 file lost coverage.
• Overall coverage decreased (-0.03%) to 91.935%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
py/dynesty/dynamicsampler.py	6	7	85.71%
<!--	Total:	22	23	95.65%	-->

Files with Coverage Reduction	New Missed Lines	%
py/dynesty/sampler.py	1	92.38%
<!--	Total:	1		-->

Totals
Change from base Build 4710600622:	-0.03%
Covered Lines:	4161
Relevant Lines:	4526

---

💛 - Coveralls

[segasai]

Now with the last commit I included the test that properly verifies that the pool is used after the restoration from the checkpoint. That helped me to discover that the first patch only fixed the case if the restoration from the checkpoint happens during the initial run. I think I'll merge this in a couple of days and probably make a new release after that.

[joshspeagle]

Good catch on this and thanks for quickly addressing the bug!

[segasai]

Okay, this thread has two more fixes of issues that I uncovered when running the tests with different seeds.

• Previously if the batch run took shorter than the checkpoint_every nothing was saved.
• When checkpointing, previously one last point before the checkpoint could have been lost leading to the differences in outputs of interrupted+restarted run vs pristine run.