Run downgrade CI on pushes to main

[ChrisRackauckas-Claude]

Summary

• run the Downgrade workflow after pushes to the repository's actual default branch, main
• leave pull-request filtering, path filtering, the reusable workflow, and the downgrade configuration unchanged

Root cause

The repository moved from master to main, but the historical branch-reference update changed only the pull_request target in Downgrade.yml. The push target remained master, so merges to the default branch did not trigger the downgrade workflow.

Impact

Merged default-branch changes will again receive strict minimum-version coverage. This is a workflow-trigger-only change; it does not alter package behavior or dependency bounds.

Validation

• actionlint -verbose .github/workflows/Downgrade.yml — exit 0, zero parse/rule errors
• julia +1.12 --startup-file=no --project=../../audit/runic_env -m Runic --check --diff . — exit 0, no formatting diff
• shell assertion over the push block — exactly one main target and no master target
• git diff --check — exit 0

Package tests were not run because the only changed datum is a GitHub Actions event filter; the package source, tests, dependency bounds, and reusable downgrade job are untouched.

Known independent failure

An exact version-only run on clean main can expose the existing SISO initialization failure. Draft #491 addresses that package behavior separately. This PR neither hides that failure nor duplicates its fix.

Process

• fetched the canonical repository and verified the authenticated fork's main is at the same commit (06464384)
• inspected the workflow's full history and the prior canonical-CI PRs before editing
• made the one-line change on a feature branch and pushed it without force

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

[ChrisRackauckas-Claude]

Implementation scratchpad for commit a051ff76:

1. Confirmed main is the repository default and compared the other push-triggered workflows, which already target main.
2. Traced Downgrade.yml history: bc036599 updated the pull-request branch from master to main but left the push branch unchanged.
3. Changed only that stale push branch; no package/test/compat or reusable-workflow settings changed.
4. Ran actionlint (zero errors), Runic (no diff), an exact trigger assertion (main once, master absent), and git diff --check (clean).
5. Did not run package tests because this event-filter edit cannot affect package execution. The known clean-main SISO failure remains isolated to test: exercise SISO initialization through ODEProblem #491 and was not skipped, silenced, or duplicated here.

[ChrisRackauckas-Claude]

CI classification after the first PR run:

• Downgrade Tests was triggered and reached the package suite, which verifies the branch-filter objective of this PR. It stopped at the independently known SISO Check failure in test/utils.jl (13 pass, 1 fail); draft test: exercise SISO initialization through ODEProblem #491 addresses that clean-main behavior and is intentionally separate.
• Core on Julia current, LTS, and prerelease all stopped at the same AccelerationSensor structurally-singular initialization fixture in test/translational.jl. That clean-main regression was independently reproduced, bisected, and fixed in draft Initialize translational acceleration sensor test #489; test: exercise SISO initialization through ODEProblem #491 is stacked on that fix.
• The downstream ModelingToolkit job is also unrelated to this workflow-only diff. Its log first loads LinearSolve v3.87.0, later resolves LinearSolve v4.3.0 [loaded: v3.87.0], and then attempts to precompile the v4 LinearSolveArnoldiMethodExt against the already-loaded v3 module, producing UndefVarError: ArnoldiMethodJL not defined in LinearSolve. The same signature is present on the unrelated Initialize translational acceleration sensor test #489 and test: exercise SISO initialization through ODEProblem #491 branches.
• Documentation, QA, Runic, Runic suggestions, and spell checking passed.

No additional change belongs in this trigger-only PR: its new main push trigger executed as intended, and it neither causes nor masks the package/base failures above.