Add Demography.to_demes().

[grahamgower]

Closes #1721.

---

WIP, but seems to be behaving itself. At least, conversion of the Demography._*_model() models look correct when plotting them with demesdraw.

[codecov]

Codecov Report

Merging #1724 (eb293bb) into main (48f2c93) will increase coverage by 0.24%.
The diff coverage is 100.00%.

❗ Current head eb293bb differs from pull request most recent head 8eac2d7. Consider uploading reports for the commit 8eac2d7 to get more accurate results

@@            Coverage Diff             @@
##             main    #1724      +/-   ##
==========================================
+ Coverage   91.14%   91.38%   +0.24%     
==========================================
  Files          20       20              
  Lines       10271    10393     +122     
  Branches     2159     2200      +41     
==========================================
+ Hits         9361     9498     +137     
+ Misses        468      454      -14     
+ Partials      442      441       -1     

Flag	Coverage Δ
C	91.38% <100.00%> (+0.24%)	⬆️
python	98.49% <100.00%> (+0.46%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
msprime/demography.py	99.41% <100.00%> (+0.04%)	⬆️
lib/msprime.c	85.65% <0.00%> (ø)
msprime/ancestry.py	98.72% <0.00%> (+<0.01%)	⬆️
msprime/provenance.py	83.49% <0.00%> (+1.94%)	⬆️
msprime/pedigrees.py	94.44% <0.00%> (+7.22%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 48f2c93...8eac2d7. Read the comment docs.

[jeromekelleher]

Looks good - is it worth letting the DemographyDebugger do some of the heavy lifting though? It's a good way of avoiding some of the multiple implementations of the event logic.

[grahamgower]

I updated to use the demography debugger for the migration matrices, and to get the deme start/end times with respect to the population's ACTIVE state. It's also desirable to check start/end times of demes based on mass migrations with proportion 1, to support legacy models, so I've left that in. Also, the big loop still iterates the events rather than the demography debugger epochs --- there's some quirk with the granularity of times returned by the demography debugger epochs such that the round tripping gives problems. I don't think it's worth the effort to try and figure that out just to shuffle a couple of lines of code.

[grahamgower]

Oh, and I updated the docstring to mention the bottleneck events can't be converted, and that legacy models could be misspecified and in general may not be convertible.

[jeromekelleher]

Looks great, thanks @grahamgower. One tiny typo spotted.

I'm generally a bit uneasy about how much logic is going in here, but I guess it can't be avoided unless we somehow modified the low-level C code to emit a demes model through the debug interface.

[grahamgower]

Typo fixed, rebased, and lightly squashed.

[jeromekelleher]

Ah, sorry - can you update the CHANGELOG with this please? This is a big feature, don't want to forget about it on release.

[grahamgower]

Changelog updated.