Fix ImplicitDiscreteSolve API docs QA

[ChrisRackauckas-Claude]

Ignore this PR until reviewed by @ChrisRackauckas.

Summary

• Point ImplicitDiscreteSolve's public-API rendering check at the OrdinaryDiffEq monorepo manual, where IDSolve is documented.
• Exclude SciMLBase-owned reexports from ImplicitDiscreteSolve's local rendering obligation while retaining the docstring check for the full public API.

Root cause

git bisect identified 12b6996e27 (Update SciMLTesting QA docs coverage (#3867)) as the first bad commit. That change moved the QA environment to SciMLTesting 2.x, whose run_qa checks rendered public API by default.

ImplicitDiscreteSolve has no package-local docs/src; its documentation lives in the monorepo's root docs/src. On Julia 1.10, the failed check therefore compared an empty documentation tree against 186 exported names: 185 SciMLBase reexports plus the package-owned IDSolve. The root manual already renders IDSolve in docs/src/misc.md.

The fix keeps the rendering assertion active for package-owned API. It only delegates SciMLBase reexports to their owner.

Validation

• ODEDIFFEQ_TEST_GROUP=QA julia +1.10 --project=lib/ImplicitDiscreteSolve -e 'using Pkg; Pkg.test()'
  • JET: 1/1 passed
  • Aqua/QA: 17/17 passed
  • Pkg.test() exited successfully
• julia +1.12 --project=<local Runic environment> -m Runic --check lib/ImplicitDiscreteSolve/test/qa/qa.jl passed with Runic 1.7.0.
• git diff --check passed.

For the bisect's good side, d7a66c0fe4 resolved SciMLTesting 1.8 and passed JET 1/1 plus Aqua 15/15. The first-bad commit reproduced the rendered-docs failure with 16 passes and 1 failure.

An additional Julia 1.12 run was blocked before this QA file loaded by a separate package-source resolution error for OrdinaryDiffEqCore's relative DiffEqBase source. That baseline is being investigated separately.

[ChrisRackauckas-Claude]

Commit 57e5d71518 contains the focused QA fix.

Investigation record:

• Reproduced the master failure with Julia 1.10: JET 1/1 passed, then API-doc QA failed with 186 unrendered names.
• Bisected from the locally passing SciMLTesting 1.8 setup to current master. First bad: 12b6996e27 (Update SciMLTesting QA docs coverage #3867).
• Verified the ownership split: 185 names are SciMLBase reexports; IDSolve is the sole package-owned export checked here and is already rendered in docs/src/misc.md.
• Kept the API-doc assertion enabled, pointed it at the root monorepo manual, and ignored only SciMLBase-owned reexports for local rendering.

Local validation after the commit:

• ODEDIFFEQ_TEST_GROUP=QA julia +1.10 --project=lib/ImplicitDiscreteSolve -e 'using Pkg; Pkg.test()': JET 1/1 and Aqua/QA 17/17 passed.
• Runic 1.7.0 check passed.
• git diff --check passed.

[ChrisRackauckas-Claude]

Relevant CI is green:

• ImplicitDiscreteSolve QA (Julia current)
• ImplicitDiscreteSolve Core (LTS, current, prerelease)
• ImplicitDiscreteSolve downgrade lane
• Documentation
• Runic format check and Runic suggestions

The overall PR remains UNSTABLE because two independent clean-master downgrade regressions reproduced in the shared matrix:

• OrdinaryDiffEqNonlinearSolve: the already-tracked MatrixOperator downgrade failure (NonlinearSolveBase 2.33.0 / safe_similar), under the existing Raise NonlinearSolveBase floor for operator Jacobians #3917 + Reconcile path-source constraints in downgrade tests .github#118 fix chain.
• OrdinaryDiffEqTaylorSeries: assigned to its separate clean-master bisect/fix investigation.

Neither failure exercises the changed ImplicitDiscreteSolve QA file.