Drop stale NaNMath 0.3 compatibility

[ChrisRackauckas-Claude]

Summary

• require NaNMath 1 in ModelingToolkitBase
• remove the stale NaNMath 0.3 line from downgrade resolution

Investigation

NaNMath 1 has been available since 2022. I changed only the ModelingToolkitBase compat entry, regenerated the exact lower-bound graph, and ran the package's existing InterfaceI group with manifest re-resolution disabled.

The current QA group also reports the repository's existing Aqua failures (45 ambiguities and 14 piracy reports). A fresh clean-base reproduction is identical, and those failures are already tracked in #4670; no Aqua change is included here.

Local verification

• Current dependency graph, GROUP=InterfaceI: 1,469 passed, 5 pre-existing broken
• Exact lower-bound graph with NaNMath 1.1.2, GROUP=InterfaceI julia +1.10 --project=lib/ModelingToolkitBase -e 'using Pkg; Pkg.test(; allow_reresolve=false)': 1,469 passed, 5 pre-existing broken
• The exact manifest SHA-256 remained cf431c15ea5c218371c4fe6f41ad96794aabdd0d407c1b62b7678ae84ea41bb1 before and after the test
• Repository-wide Runic check: passed
• Clean-base QA reproduction: 54/54 JET checks; Aqua 9 passed, 2 failed, exactly matching the candidate

Ignore this PR until it has been reviewed by @ChrisRackauckas.

[ChrisRackauckas-Claude]

Investigation log:

1. Rebased the one-line compat change onto current origin/master (d2e576d2).
2. Regenerated the exact lower-bound environment; it selected NaNMath 1.1.2.
3. Locked that manifest and ran the existing InterfaceI group with allow_reresolve=false.
4. Observed 1,469 passes and 5 pre-existing broken tests in 20m36.3s.
5. Verified the manifest SHA-256 was unchanged before and after the test.
6. Re-ran the repository-wide Runic check successfully.
7. Independently reproduced the QA failures on a fresh clean base and matched all 45 ambiguity and 14 piracy reports; existing issue QA: track pre-existing Aqua/JET/ExplicitImports findings (root + ModelingToolkitBase) #4670 tracks them.

The branch contains only commit 3b874ad785, with the required co-author trailer.

[ChrisRackauckas-Claude]

CI triage for 3b874ad7:

• The target Downgrade Tests - InterfaceI lane passed.
• The lib/ModelingToolkitBase downgrade failure matches the clean-master result already compared on Drop stale DiffRules 0.1 compatibility #4793: 7,235 passed / 3 errored / 4 broken in the same MIRK optimal-control tests. The underlying interpolation fix is BoundaryValueDiffEq #542.
• ModelingToolkitBase QA is the clean-master Aqua ambiguity/piracy baseline tracked by #4670.
• All three sublibrary InterfaceII jobs and the root QA job have GitHub annotations stating that their self-hosted runners lost communication.
• The Catalyst Modeling and ModelingToolkitStandardLibrary failures are also reproduced clean-base failures, independent of this one-line NaNMath compat edit.

The requested stale-major removal is green; the inspected red lanes are clean-base failures or runner infrastructure rather than regressions from this branch.