Better assy solver implementation

[adam-urbanczyk]

This PR implements a new assy solver based on Casadi (for autodiff) and IPOPT (actual optimizer, called via casadi).

[adam-urbanczyk]

@lorenzncode not finished yet, but you could already take a look.

[codecov]

Codecov Report

Merging #1063 (2988163) into master (803a05e) will increase coverage by 0.04%.
The diff coverage is 98.01%.

❗ Current head 2988163 differs from pull request most recent head 407bec5. Consider uploading reports for the commit 407bec5 to get more accurate results

@@            Coverage Diff             @@
##           master    #1063      +/-   ##
==========================================
+ Coverage   96.32%   96.37%   +0.04%     
==========================================
  Files          40       40              
  Lines        9427     9496      +69     
  Branches     1251     1256       +5     
==========================================
+ Hits         9081     9152      +71     
+ Misses        204      202       -2     
  Partials      142      142              

Impacted Files	Coverage Δ
cadquery/assembly.py	96.55% <92.85%> (+1.50%)	⬆️
cadquery/occ_impl/solver.py	96.20% <98.27%> (+0.30%)	⬆️
cadquery/occ_impl/geom.py	99.58% <100.00%> (+<0.01%)	⬆️
tests/test_assembly.py	100.00% <100.00%> (ø)

📣 Codecov can now indicate which changes are the most critical in Pull Requests. Learn more

[lorenzncode]

What a big effort the new implementation is.

I tested the constraints and did not find any significant problem.

• FixedRotation: I would think gp_Extrinsic_XYZ might be a better default.

• Could improve handling of a single unary constraint such as the following.

  assy.constrain("b1", "FixedRotation", (45,0,45))

  Currently it is allowed, however, ends with Number of Iterations....: 0

  The workaround is to add a dummy object and "Fixed" constraint. Of course a realistic assembly is expected to have more constraints but a new user might experiment like this. Either require a "Fixed" constraint or allow a single unary constraint without locking first entity.

[adam-urbanczyk]

Thanks @lorenzncode . I changed to extrinsic and now it is explicitly disallowed to have only one unary constraint (or only one constrained entity to be exact).

[lorenzncode]

The print output to stdout is useful. Perhaps an option can be exposed to reduce verbosity if the user chooses (not necessary for this PR can open a separate issue).

assy.solve(options={"verbose": 0})

(key "verbose", "verbosity", "print_level" or similar)

An options dict may be potentially useful to pass future options as well to control other algorithm settings or preferences (passed through to ConstraintSolver).

[adam-urbanczyk]

Let's open another issue for this. TBH I was thinking more in the direction of removing it in the near feature (once the solver is more established.

[adam-urbanczyk]

OK I think it is good enough now @lorenzncode @jmwright

[jmwright]

+1 to merge @adam-urbanczyk @lorenzncode

[lorenzncode]

Thanks @adam-urbanczyk!