Drop stale DiffRules 0.1 compatibility

[ChrisRackauckas-Claude]

Summary

• require DiffRules 1 in ModelingToolkitBase
• remove the stale DiffRules 0.1 compatibility line from downgrade resolution

Investigation

The ModelingToolkitBase lower-bound graph could still select DiffRules 0.1 even though the package's current Symbolics and NaNMath constraints require the DiffRules 1 line. The branch changes only that compatibility entry.

Local verification

• Rebased head bb9b3fdcad onto current upstream master at f6c5c80da6.
• Current root graph on Julia 1.12, GROUP=InterfaceI: 1,475 passed and 3 pre-existing broken; Pkg.test() exited 0 in 44m12.6s.
• Fresh ModelingToolkitBase lower-bound graph on Julia 1.10 selected ModelingToolkitBase 1.54.2 and DiffRules 1.10.0.
• Locked lower-bound graph, GROUP=InterfaceI with Pkg.test(; allow_reresolve=false): 1,495 passed and 5 pre-existing broken; exit 0 in 20m39.3s.
• The exact manifest SHA-256 remained d0c002a3661863d019bbb9107f7c795b00aa7561a0efd1272f3a018ae0c0e075 before and after the test.
• The exact compat assertion and git diff --check passed.
• Runic 1.7 checked 221 Julia files. The 220 files other than lib/ModelingToolkitBase/src/systems/abstractsystem.jl passed; that untouched current-master failure was reproduced, bisected, locally fixed, and opened separately as draft PR Fix Runic indentation in parse_variable #4801.

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 DiffRules 1.10.0.
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 20m05.9s.
5. Verified the manifest SHA-256 was unchanged before and after the test.
6. Confirmed the repository-wide Runic check was clean.

The branch contains only commit 0bfc45a74f, with the required co-author trailer.

[ChrisRackauckas-Claude]

CI triage for 0bfc45a:

• The target root downgrade lane, Downgrade Tests - InterfaceI, passed.
• The lib/ModelingToolkitBase downgrade failure is identical on clean master: both jobs report 7,235 passed / 3 errored / 4 broken, with the same Lotka–Volterra, cost-function-compilation, and parameter-estimation errors. PR job, clean-master job.
• The sublibrary InterfaceII failures reproduce on an unmodified clean checkout under Julia 1.12 with the same 7,235 / 3 / 4 result. The underlying MIRK dual-interpolation fix is already isolated in BoundaryValueDiffEq #542; the Julia 1 and pre jobs additionally lost their self-hosted runners.
• The sublibrary QA result (9 passed / 2 failed: Aqua ambiguity and piracy) also reproduces on clean master and is tracked in #4670.
• The root QA job did not report a test failure; its sole annotation says the self-hosted runner lost communication with GitHub.

Thus the requested DiffRules minimum is green, and every inspected red ModelingToolkit lane is either a clean-master failure or runner infrastructure rather than a regression from this one-line compat change.

[ChrisRackauckas-Claude]

Follow-up classification for the three downstream reds not listed above: both Catalyst Modeling lanes and the ModelingToolkitStandardLibrary Core lane were independently reproduced against current clean master during the later #4773 audit. The Catalyst baseline is 20 passed / 1 parent_sys error and is addressed by Catalyst #1511; the MTSL baseline is 15 passed / 1 AccelerationSensor structural-singularity error and is addressed by ModelingToolkitStandardLibrary #495. Full job links and evidence are recorded in #4773 (comment). This PR changes only the DiffRules compat entry, and its requested root downgrade lane passed, so no branch change is warranted.

[ChrisRackauckas-Claude]

Rebase validation for exact head fa64c5cb4906e5a7e6dfb0987ad1a5c623196ac3:

1. Refetched upstream and the fork, verified the remote branch still contained only the original one-line commit, and rebased it onto current master (80edc5dc9a831de7ae229b069134d03025991b41).
2. Ran the official root InterfaceI group on Julia 1.12: 1,475 passed / 3 pre-existing broken; Pkg.test() exited 0 in 44m53.5s.
3. Regenerated the ModelingToolkitBase lower-bound graph on Julia 1.10. It selected ModelingToolkitBase 1.54.2 and DiffRules 1.10.0.
4. Locked that graph and ran InterfaceI with allow_reresolve=false: 1,495 passed / 5 pre-existing broken; exit 0 in 20m21.2s.
5. Verified the manifest SHA-256 stayed d0c002a3661863d019bbb9107f7c795b00aa7561a0efd1272f3a018ae0c0e075 before and after testing. The compat assertion and git diff --check also passed.
6. Runic 1.7 passed 220 of 221 Julia files. Its sole finding is in untouched lib/ModelingToolkitBase/src/systems/abstractsystem.jl at the current-master parent; the required clean-base reproduction/bisect is running separately.

The branch remains a one-line Project.toml compatibility change with the required co-author trailer.

[ChrisRackauckas-Claude]

Fresh validation for exact head bb9b3fdcad3d62539b15b270a3c67cb211c2a08d:

1. Refetched upstream and the fork, confirmed the fork branch still contained only the prior agent-owned commit, and rebased onto current master (f6c5c80da6cfad33b152033604a5c57b4548814e).
2. Ran the official root InterfaceI group on Julia 1.12: 1,475 passed / 3 pre-existing broken; Pkg.test() exited 0 in 44m12.6s.
3. Regenerated the ModelingToolkitBase lower-bound graph on Julia 1.10. It selected ModelingToolkitBase 1.54.2 and DiffRules 1.10.0.
4. Locked that graph and ran InterfaceI with allow_reresolve=false: 1,495 passed / 5 pre-existing broken; exit 0 in 20m39.3s.
5. Verified the manifest SHA-256 remained d0c002a3661863d019bbb9107f7c795b00aa7561a0efd1272f3a018ae0c0e075 after testing. Both worktrees were clean, the compat assertion and git diff --check passed, and the branch remains one file / one insertion / one deletion / one co-authored commit.
6. The sole Runic finding is unchanged from this branch's current-master parent. It was separately reproduced, bisected to 3ae5bc97d8, fixed, and validated in draft PR Fix Runic indentation in parse_variable #4801.

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

[ChrisRackauckas-Claude]

Exact-head CI classification for bb9b3fdcad3d62539b15b270a3c67cb211c2a08d:

• Final hosted tally: 79 success / 10 failure / 3 skipped / 1 cancelled. The root Downgrade check and all InterfaceI checks succeeded.
• Fresh local validation at this exact head passed the current root InterfaceI group (1,475 pass / 3 pre-existing broken / 1,478 total) and the locked ModelingToolkitBase lower-bound graph (1,495 pass / 5 pre-existing broken / 1,500 total). The lower-bound manifest hash remained unchanged.
• The ModelingToolkitBase downgrade job lost its self-hosted runner. A targeted rerun was attempted, but GitHub returned HTTP 403 because admin rights are required.
• The remaining runner losses/timeouts and substantive test errors were reproduced on the exact clean parent f6c5c80da6cfad33b152033604a5c57b4548814e. The clean-parent check set had 10 failures; this head does not introduce a deterministic product failure.
• The Catalyst Hybrid variance result was reproduced exactly on both the clean parent and this head with byte-equivalent dependency graphs. A focused statistical stabilization is now draft PR Catalyst #1519: it keeps the tolerance unchanged, increases independent Brownian/jump trials, passes the formerly failing RNG (4,218/4,218), and passes the official current Hybrid group (4,218/4,218).
• The clean-parent Runic failure is independently fixed by draft PR #4801, whose exact head has Runic green and no head-only failures.

The compatibility change remains a one-line removal of unreachable DiffRules 0.1 support and the PR remains a draft for review.