Make the tests independent of polymake's convex hull backend

[fingolfin]

Fixes #18. Stacked on #27.

The failure was caused by different convex hull backends producing different (but equally valid) results.

To fix this, canonicalise before comparing:

• FACES -- sort by dimension, then lexicographically, so the expected value
  still reads like a face lattice.

• FACETS -- scale each row to a primitive integer vector, then sort. Signs
  are preserved: every differing pair I checked differs by a positive factor,
  and facet orientation is meaningful, so normalising it away would discard
  real information.

• VERTICES_IN_FACETS -- sort. The VERTICES order is stable (that
  expectation already passed).

• ADJACENCY -- checked structurally instead. It is the covering relation of
  FACES, and the same call returns both in the same node order, so no
  relabelling is needed:

  gap> List(adj, Set) = CoversOfFaceList(faces);
  true

  Canonicalising it by relabelling would have produced a 46-line literal that
  nobody could review. A degree-sequence check is included alongside. Note the
  relation is directed upward only, so a symmetry check would be wrong.

The suite no longer rewrites checked-in files

polymaking.tst pointed the data directory at the package's own tst/ and
read pplane.poly from there. polymake upgrades old-format files in place, so
running the tests dirtied the working tree and made a second run test a
different input. This happens on the CI runner too, not just on Gentoo:

polymake: upgrading /home/runner/work/polymaking/polymaking/tst/pplane.poly from old plain file format

Now copied to a temporary directory first, with git diff --exit-code in CI to
catch a regression.

Testing

Set POLYMAKING_CHULL to pick a backend:

POLYMAKING_CHULL=cdd gap tst/testall.g

0 failures under ppl, cdd, lrs, beneath_beyond and the default, and
git status is clean afterwards. CI gains cdd and beneath_beyond jobs;
beneath_beyond is built into polymake so it is always available, and cdd is
what Debian and Gentoo builds tend to use.

I used an environment variable rather than writing gap.ini because I could
not confirm that gap-actions/run-pkg-tests reads gap.ini; an env var is
immune to how CI invokes GAP, and makes local reproduction a single line.

While here, repaired visual.tst, which had rotted while excluded from the
suite -- it expected no output from LoadPackage and had a stray trailing
space. It passes when run manually now.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 72.23%. Comparing base (47eb6ec) to head (c02c5f9).

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #28      +/-   ##
==========================================
+ Coverage   72.00%   72.23%   +0.23%     
==========================================
  Files          15       15              
  Lines         868      868              
==========================================
+ Hits          625      627       +2     
+ Misses        243      241       -2     

see 1 file with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.