Replace DataStructures.jl with BinaryHeaps.jl to cut invalidations

[ChrisRackauckas-Claude]

Summary

Swaps the DataStructures dependency for the new lightweight SciML/BinaryHeaps.jl, which contains the same BinaryHeap / BinaryMinHeap / FasterForward code extracted directly from DataStructures.jl (same upstream source, MIT-licensed, original copyright preserved).

Depends on: SciML/BinaryHeaps.jl#1 being merged and v1.0.0 registered in General.

Motivation

Loading DataStructures.jl causes 6 method invalidation trees from code that is completely unrelated to heap functionality:

Tree 1: kwcall(...sort!...LittleDict...) @ OrderedCollections
Tree 2: get!(h::OrderedDict...)          @ OrderedCollections
Tree 3: merge!(ct::Accumulator...)       @ DataStructures
Tree 4: getkey(dd::LittleDict...)        @ OrderedCollections
Tree 5: Base.IteratorSize(...SortedDict/SortedMultiDict/SortedSet...) @ DataStructures
Tree 6: merge(m::SortedMultiDict...)     @ DataStructures

OrdinaryDiffEq only uses BinaryHeap{T}(FasterForward()), BinaryMinHeap{T}, push!, pop!, first, isempty, empty!, sizehint! — a tiny slice of DataStructures. Pulling in the full package costs 6 invalidation trees and their associated first-call latency for zero benefit.

BinaryHeaps.jl provides the exact same API with zero invalidations.

Changes

• OrdinaryDiffEqCore: using DataStructures: BinaryHeap, FasterForward → using BinaryHeaps: BinaryHeap, FasterForward; three DataStructures.FasterForward() → FasterForward() in solve.jl.
• DelayDiffEq: using DataStructures: BinaryMinHeap → using BinaryHeaps: BinaryMinHeap.
• StochasticDiffEqCore: removed the DataStructures import (it was unused — nothing from the package was referenced).
• Project.tomls: swap dep/compat entries accordingly; top-level OrdinaryDiffEq/Project.toml drops the unused DataStructures entry.

Diff is 8 files, +10 / -15 lines.

Verification

Locally tested with BinaryHeaps.jl dev-installed:

• OrdinaryDiffEqTsit5 loads and runs with tstops/saveat (both use BinaryHeap{T}(FasterForward())):

  sol = solve(prob, Tsit5(); tstops=[0.3, 0.6], saveat=[0.25, 0.75])  # ✓ ok

• DelayDiffEq loads and runs a DDE with a constant lag (uses BinaryMinHeap for propagated discontinuities):

  sol = solve(DDEProblem(...; constant_lags=[1.0]), MethodOfSteps(Tsit5()))  # ✓ ok

Test plan

• All touched subpackages load without errors
• Basic ODE solve with tstops/saveat works
• Basic DDE solve with constant_lags works
• Full CI passes (awaits BinaryHeaps.jl registration)

🤖 Generated with Claude Code