Require SciMLTesting 1.2 for grouped test harness

[ChrisRackauckas-Claude]

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

This is the rebased successor to #4790. GitHub would not reopen that closed PR after its head branch was force-updated onto current master.

Summary

• raise the root SciMLTesting test floor from 1 to 1.2
• retain the newly added SciMLTesting 2.1 compatibility from Allow SciMLTesting v2.1 in QA envs #4758
• align the declared 1.x minimum with the folder-based run_tests API adopted in d647d61ac4

Why #4758 does not supersede this floor

Current master still declares SciMLTesting = "1, 2.1". Its root harness calls grouped run_tests without a core keyword at test/runtests.jl:9. SciMLTesting 1.0.0 through 1.1.1 require that keyword, so every still-allowed 1.x resolution below 1.2 fails before any package test runs.

An exact SciMLTesting 1.0 control on current master (fb3706f352) reproduces:

UndefKeywordError: keyword argument `core` not assigned
  @ test/runtests.jl:9

The 2.x QA migration makes current resolution work, but it does not repair the declared 1.x compatibility range. This PR changes that range from 1, 2.1 to 1.2, 2.1.

Local validation after rebasing onto current master

• Julia 1.10.11, SciMLTesting pinned exactly to 1.2.0, official GROUP=SymbolicIndexingInterface: 1,964/1,964 passed
• Julia 1.12.6, resolved SciMLTesting 2.4.0, same official group: 1,964/1,964 passed
• clean exact SciMLTesting 1.0 control: reproduced the core keyword startup failure above
• git diff --check: passed
• repository-wide Runic 1.7 exits 1 on the independent clean-master indentation defect fixed by Fix Runic indentation in parse_variable #4801; this PR changes only Project.toml

Earlier boundary probes also confirmed 1.0.0, 1.1.0, and 1.1.1 fail at harness startup while 1.2.0 is the first compatible release.

[ChrisRackauckas-Claude]

Validation scratchpad for b55f739adc228cc3e31439e2feb600855eb1428f on base fb3706f35208922b634699c3bf325d5721564f07:

• Julia 1.10.11 with SciMLTesting = "=1.2.0", using the repository's official GROUP=SymbolicIndexingInterface entrypoint: 1,964/1,964 passed in 4m01.6s after precompilation. The resolver selected SciMLTesting 1.2.0. Raw-log SHA-256: 3689a78426abb9e8e29afd1321890943ce227b5210a4608920a7dd8014675139.
• Julia 1.12.6 with the committed compat and resolved SciMLTesting 2.4.0, using the same official group: 1,964/1,964 passed in 9m02.2s. Raw-log SHA-256: 3cfdfce13e2472dd521a33927162687772783038c6ad246f115c8c20e0941bbd.
• Exact SciMLTesting 1.0 on clean current master fails before package tests with UndefKeywordError: keyword argument core not assigned at test/runtests.jl:9. Raw-log SHA-256: 007a44a1cae13272775f24afeed798a9e8a1146372d5eaac2c72828c32edea37.
• git diff --check passed and the feature worktree remained clean.
• Repository-wide Runic 1.7 exits 1 on the clean-master indentation defect already fixed by Fix Runic indentation in parse_variable #4801. This one-line PR changes only TOML compat.

No failing test was skipped, silenced, or loosened.

[ChrisRackauckas-Claude]

Final exact-head CI classification for b55f739adc228cc3e31439e2feb600855eb1428f against unchanged base fb3706f35208922b634699c3bf325d5721564f07:

• All 67 checks are terminal: 57 success / 6 failure / 4 skipped.
• The target Downgrade / Downgrade Tests - InterfaceI check passed.
• All Julia 1/LTS/prerelease InterfaceI and InterfaceII checks passed, as did every other completed root functional lane.
• Root QA and the ModelingToolkitBase downgrade job each ended only because their self-hosted runner lost communication; both check annotations say so and contain no test assertion failure.
• Runic reproduces the clean-master indentation defect fixed by Fix Runic indentation in parse_variable #4801. The exact log diff is the parse_variable indentation change from that PR.
• One Catalyst downstream job reaches the known private-AffectSystem parent_sys failure; the other passes Symbolic Stoichiometry 30/30 and reaches the seven Conservation Laws unexpected passes fixed by Catalyst Add formatter #1512.
• ModelingToolkitStandardLibrary reaches the known clean-base AccelerationSensor structural-singularity fixture fixed by MTSL Restructure tutorial section of the documentation #489.

None of the six failures is caused by the one-line SciMLTesting compat change. The branch remains one commit/one file, is zero commits behind current master, and the exact SciMLTesting 1.2.0 and current 2.4.0 local groups each passed 1,964/1,964 as recorded above.

No test was skipped, silenced, or loosened.

[ChrisRackauckas]

Dropping all pre 2.4