Use SciMLTesting public API docs QA

[ChrisRackauckas-Claude]

Ignore this PR until reviewed by @ChrisRackauckas.

Summary

• Update the package QA/docs coverage branch for SciMLTesting v2.1 public API documentation checks.
• Remove or avoid package self [sources] path entries in the touched QA/docs environments.
• Keep this draft while the fleet-wide public/exported API documentation audit is reviewed.

Validation

• Branch worktree was clean before opening this PR.
• Local validation was run by the worker before push where applicable; this PR-opening pass did not rerun the full package test suite.

[ChrisRackauckas-Claude]

CI triage result: the remaining red check looks runner/infrastructure-specific, not caused by this PR's SciMLTesting QA change.

Evidence:

• Current head d6f7910: GPU job failed on V100 runner self-hosted-gpu-v100-8vcpu-16gb-zjbnn-runner-s4hck with 36 CUDNNError: CUDNN_STATUS_EXECUTION_FAILED_CUDART errors in test/layers_tests.jl: https://github.com/SciML/DeepEquilibriumNetworks.jl/actions/runs/29149809271/job/86537393051
• Earlier branch commit 524cb36, before the SciMLTesting compat change (9f0aac9), already failed the same GPU lane on a V100 runner with the same 36 cuDNN errors while resolving SciMLTesting v1.8.0: https://github.com/SciML/DeepEquilibriumNetworks.jl/actions/runs/29113750846/job/86432077386
• Base main commit 6ac9651 had the GPU lane green on a T4 runner: https://github.com/SciML/DeepEquilibriumNetworks.jl/actions/runs/29114758096/job/86435378910
• This PR's current diff is only Project.toml and test/qa/Project.toml; QA/Core/docs/Runic/downgrade checks are green on the current head.

Local verification of the touched QA path passed:
GROUP=QA timeout 3600 ~/.juliaup/bin/julia +release --project=. -e 'using Pkg; Pkg.test()'
Result: QA | 19 passed / 19 total.

Recommendation: rerun the GPU job or route/fix the V100 runner CUDA/cuDNN environment. I do not see a focused PR code change to make here.