Fix Julia 1.10 support and add downgrade CI

[ChrisRackauckas-Claude]

Summary

This PR restores Julia 1.10 (LTS) support and adds a downgrade-compatibility CI workflow for the three workspace submodules.

1. Julia 1.10 code fix (DiscreteMeasures)

DiscreteMeasures.jl/src/DiscreteMeasures.jl used the two-argument form allequal(length, x), which only exists on Julia >= 1.11. On Julia 1.10 (the SciML-required LTS) this throws a MethodError, so the package cannot even load/construct a DiscreteMeasure. Replaced with allequal(length.(x)), which is semantically identical (all support points share the same length) and works on 1.10.

Verified on the 1.10 binary: the two-arg form MethodErrors; allequal(length.(x)) returns the same result (true for equal-length / scalar vectors, false otherwise).

2. CanonicalMoments compat floor fix

IntervalArithmetic: "0.16 - 0.20, =0.20.9" -> "0.18 - 0.20, =0.20.9". IntervalArithmetic 0.16-0.17 require RecipesBase <= 0.7 (and pull ConstructionBase down), which is unsatisfiable together with this package's Symbolics floor. 0.18 is the lowest IntervalArithmetic that resolves and passes the suite. The disjoint =0.20.9 range is preserved; no allowed version sits below the verified floor.

DiscreteMeasures and OUQBase floors were already healthy on 1.10 (no compat change needed).

3. Downgrade CI

Added .github/workflows/Downgrade.yml: a matrix over [CanonicalMoments.jl, DiscreteMeasures.jl, OUQBase.jl] calling SciML/.github/.github/workflows/downgrade.yml@v1 with julia-version: "1.10", project: ${{ matrix.project }}, projects: ${{ matrix.project }}, secrets: inherit. Mirrors the existing Tests.yml matrix shape.

Depends on SciML/.github#67 (adds the project input to the reusable downgrade.yml so it can build/test a workspace submodule) + a v1 retag. Until that merges and v1 is retagged, the project: input is ignored by the pinned @v1 reusable workflow.

Local verification (Julia 1.10, isolated depot, compat pinned to floors)

All three submodules pass the downgrade suite at their compat floors on Julia 1.10:

DiscreteMeasures (IntervalArithmetic 0.16.0):
Test Summary:          | Pass  Total
DiscreteMeasure        |    8      8
ProductDiscreteMeasure |   15     15
Expectation            |    3      3
Measure Restrictions   |   19     19
     Testing DiscreteMeasures tests passed

CanonicalMoments (IntervalArithmetic 0.18.0, Symbolics 6.29.2, Reexport 1.2.2):
Test Summary:               | Pass  Total
Orthogonal Polynomial Roots |   11     11
Moment Sequence             |  260    260
     Testing CanonicalMoments tests passed

OUQBase (ModelingToolkit 9.69.0, SciMLBase 2.89.1, Symbolics 6.39.1, JuMP 1.23.6,
         DomainSets 0.7.15, Polynomials 4.0.13, OrderedCollections 1.7.0,
         NaNMath 1.1.3, Reexport 1.2.2):
     Testing Running tests...
Optimization stopped after 101 steps
Best candidate found: [0.376915, 0.851782]
Fitness: 0.198058667
     Testing OUQBase tests passed

Ignore until reviewed by @ChrisRackauckas.

🤖 Generated with Claude Code

[ChrisRackauckas-Claude]

Downgrade CI fix for CanonicalMoments + OUQBase

Root cause (both failures, identical): julia-downgrade-compat resolves each workspace submodule in isolation, and the submodules declared their sibling submodules as plain [deps] with no way to locate them. Since DiscreteMeasures/CanonicalMoments are unregistered local path deps, the action's standalone resolver failed with:

[ Info: Running resolver on CanonicalMoments.jl with --min=@deps
ERROR: LoadError: unknown package UUID: 7766d772-2108-41ee-a4bd-11c51440a39b   # DiscreteMeasures

DiscreteMeasures passes because it has no sibling deps.

Fix: declare sibling submodules via [sources] path entries. The downgrade action (and SciML/.github tests.yml@v1) explicitly handle [sources] path deps — they are removed for registry resolution and re-added to the manifest as path deps. This is the canonical workspace mechanism; the latest-CI (Julia 1.12) tests are unaffected.

Once resolution worked, two genuine floor bugs surfaced and were fixed (lowest working bounds):

CanonicalMoments (deps had no compat at all → floated to ancient versions):

• Polynomials: none → "4" (resolved 0.5.0 lacked the AbstractPolynomial export the package uses → UndefVarError: AbstractPolynomial not defined)
• RecurrenceRelationships: none → "0.2"
• added julia="1.10", LinearAlgebra, Statistics, DiscreteMeasures floors

OUQBase:

• SciMLBase: 2.89.1 → 2.153. The 2.89.1 floor held Optimization down to 3.25.1, but the test stack floated OptimizationBBO to a version calling deduce_retcode that the old Optimization lacked → UndefVarError: deduce_retcode not defined during the FloodProblem solve. Raising the SciMLBase floor lets OptimizationBBO/OptimizationBase/Optimization resolve consistently (matches the passing latest set).

Local downgrade suite (Julia 1.10, deps-mode, allow_reresolve=true): PASS

• CanonicalMoments — Orthogonal Polynomial Roots 11/11, Moment Sequence 260/260 (Polynomials 4.0.0, RecurrenceRelationships 0.2.0)
• OUQBase — FloodProblem solve pipeline completes (SciMLBase 2.153.0, OptimizationBBO 0.4.5, OptimizationBase 4.2.0); previously errored on deduce_retcode
• DiscreteMeasures — already green, unchanged

Downgrade.yml already sets allow-reresolve: true for all three matrix entries (single matrix-templated with: block), so no workflow change was needed there.

Ignore until reviewed by @ChrisRackauckas.

🤖 Generated with Claude Code

[ChrisRackauckas-Claude]

Switched downgrade to strict allow-reresolve: false

Per @ChrisRackauckas, the Downgrade.yml caller now uses the strict SciML default allow-reresolve: false instead of true. This is the rigorous downgrade: julia-downgrade-compat@v2 (mode deps) resolves each workspace submodule's [deps] to their floors with transitive deps pinned to LATEST, and Pkg.test must instantiate that exact set with no reconciliation.

Local verification (Julia 1.10, deps-mode, Pkg.test(; allow_reresolve=false), per matrix submodule)

Reproduced the exact CI steps locally (ran the action's downgrade.jl resolver against each submodule, then strict test). All three PASS at the existing floors — no additional floor raises were needed beyond those already in this PR.

Submodule	Result
DiscreteMeasures.jl	PASS — 45/45 (DiscreteMeasure 8, ProductDiscreteMeasure 15, Expectation 3, Measure Restrictions 19)
CanonicalMoments.jl	PASS — Orthogonal Polynomial Roots 11/11, Moment Sequence 260/260
OUQBase.jl	PASS — FloodProblem OptimizationBBO solve pipeline completes (best candidate [0.344441, 0.831005], fitness 0.1927)

The floors added earlier in this PR (CanonicalMoments: Polynomials = "4", RecurrenceRelationships = "0.2", sibling [sources]; OUQBase: SciMLBase = "2.153", sibling [sources]) are exactly the floors that make the strict floor-direct / latest-transitive set resolve and test-pass, so flipping to false is clean.

Only .github/workflows/Downgrade.yml changed in this commit (allow-reresolve: true → false).

Ignore until reviewed by @ChrisRackauckas.

🤖 Generated with Claude Code

[ChrisRackauckas-Claude]

Follow-up: the lib/ monorepo restructure + centralized sublibrary CI uniformization (which was scoped as an extension of this PR) is in #10, since this PR was already merged. All fixes from this PR (Julia 1.10, [sources], verified [compat] floors) are in main and preserved by #10.