test: always-run canary smoke set spanning physics modules

[sbryngelson]

Addresses the selection-robustness half of #1587 / #1589 (the fix itself is #1588).

Why

#1556's silent viscous-GPU regression merged green because CI's coverage-based test
selection (--select-enforce) + sharding never ran a viscous case on the AMD-flang lane
for that diff — even though the suite already has very loud viscous tests (Re = 1e-4). The
gap is selection, not loudness or missing tests: any feature that fails silently (viscosity,
surface tension, bubbles, MHD, elasticity, chemistry, IBM, …) can slip through if its tests
aren't in the selected/sharded subset.

What

A curated always-run "canary" smoke set — one cheap, feature-dominant existing regression
case per major physics module — tagged so select_tests always includes them, regardless of
coverage overlap. A silent regression in any covered module then trips on every PR/lane.

• test/case.py: add a canary: bool field to TestCase / TestCaseBuilder.
• test/coverage.py: select_tests() always keeps canary cases (never rung-7 skipped).
• test/cases.py: _CANARY_TRACES (12 traces, one per module) tagged in list_cases(), with
  a validation guard so a renamed/removed trace is a loud error, not a silently dropped
  canary.
• test/test_coverage_unit.py: unit test that a canary survives a coverage-miss that would
  otherwise skip it. (44/44 unit tests pass.)

Tagging existing cases → no new golden files. Net +60/−3 across 4 toolchain files.

The set (cheapest 1D/2D variant per module):

module	canary trace
viscosity (m_viscous)	1D -> 1 Fluid(s) -> Viscous
non-Newtonian (m_hb_function)	1D -> 1 Fluid(s) -> Non-Newtonian -> nn=0.5
surface tension (m_surface_tension)	2D -> 2 Fluid(s) -> capillary=T -> model_eqns=3
bubbles EE/QBMM (m_qbmm)	1D -> Bubbles -> QBMM
bubbles EL (m_bubbles_EL)	2D -> Lagrange Bubbles -> One-way Coupling
MHD + HLLD (m_riemann_solver_hlld)	1D -> MHD -> HLLD
hypoelasticity (m_hypoelastic)	1D -> Hypoelasticity -> 1 Fluid(s)
chemistry	1D -> Chemistry -> Perfect Reactor
IBM (m_ibm)	2D -> 1 Fluid(s) -> IBM -> Circle -> slip
phase change / 6-eq (m_pressure_relaxation)	1D -> Phase Change model 5 -> 2 Fluid(s) -> model equation -> 3
acoustic source (m_acoustic_src)	1D -> Acoustic Source -> Sine -> Frequency
body forces (m_body_forces)	1D -> Bodyforces

Verification

• ./mfc.sh test -l (runs the canary validation) green; 12/12 cases tagged.
• ./mfc.sh precheck green (all 7 checks, incl. toolchain unit tests).
• The canaries are existing, already-passing cases; this PR's CI exercises them across all
  lanes — the real end-to-end check. Worth a glance at total added CI time; the set is easy to
  trim if needed (it's one frozenset).

Notes / follow-ups

• Hyperelasticity has no cheap dedicated regression case (only example-driven) — a genuine
  coverage gap, tracked for follow-up rather than blocking here.
• Sharding partitions before selection, so each canary runs in exactly one shard; every shard
  is its own CI job and all run, so canaries are always executed.
• Remaining Harden AMD gpu-omp CI coverage + lint against silent declare-target regressions (follow-up to #1588) #1589 items (metamorphic feature-on≠off check; GPU_DECLARE static-array precheck
  rule) are separate.

[Copilot]

Pull request overview

This PR hardens the Python toolchain’s coverage-based test selection by introducing an always-run “canary” smoke set: a curated subset of existing regression cases (one per major physics module) that should be executed on every lane even when normal coverage overlap would skip them.

Changes:

• Add a canary: bool flag to test case definitions/builders and tag a curated set of traces as canaries.
• Update coverage-based selection to never skip canary cases during per-test selection.
• Add a unit test ensuring a canary case survives a coverage-miss scenario that would otherwise skip it.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
toolchain/mfc/test/test_coverage_unit.py	Adds a unit test validating that canary-tagged cases are selected even when their coverage does not overlap changes.
toolchain/mfc/test/coverage.py	Ensures canary cases are always included during per-test (rungs 5–7) selection.
toolchain/mfc/test/cases.py	Defines the canary trace set and validates/tag them in list_cases() (loud failure if a canary trace is missing).
toolchain/mfc/test/case.py	Adds canary to TestCase/TestCaseBuilder and propagates it when building runnable cases.

[sbryngelson]

Good catch — fixed in a335da2 (the convergence branch now forwards canary=self.canary too). For the record it isn't a live bug: no convergence case is in _CANARY_TRACES, and select_tests reads canary off the TestCaseBuilder before to_case() runs (selection happens prior to the [_.to_case() for _ in cases] conversion). But you're right that it's an inconsistent footgun, so the branches now match.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 60.94%. Comparing base (ac5774e) to head (a335da2).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1590   +/-   ##
=======================================
  Coverage   60.94%   60.94%           
=======================================
  Files          82       82           
  Lines       19922    19922           
  Branches     2924     2924           
=======================================
  Hits        12141    12141           
  Misses       5805     5805           
  Partials     1976     1976           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

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