Fix downgrade workflow trigger for main

[ChrisRackauckas-Claude]

Please ignore this PR until it has been reviewed by @ChrisRackauckas.

Summary

Point both downgrade workflow triggers at the repository's actual default branch, main, instead of the retired master name.

The stale pull-request and push filters meant the downgrade workflow did not run for normal main development.

Local verification

• Reproduced the deployed downgrade workflow locally at commit 5fdbe2bd using Julia 1.10.11 and julia-downgrade-compat fab1def: 1,297/1,297 assertions passed across Core, AD, and Enzyme coverage.
• Pkg.build() passed in the exact floor environment.
• actionlint .github/workflows/Downgrade.yml passed.
• Runic 1.7 passed for all 15 Julia files.
• git diff --check passed.

[ChrisRackauckas-Claude]

The red downgrade check is an infrastructure failure, not a test result. GitHub's job record shows that resolution and julia-buildpkg both succeeded, then the self-hosted runner lost communication while julia-runtest was still marked in_progress; no test failure or completed test step was recorded.

I attempted to rerun only the failed job, but GitHub returned 403 Must have admin rights to Repository. The exact workflow was already reproduced locally at this head with 1,297/1,297 assertions passing, as documented in the PR body.

[ChrisRackauckas-Claude]

CI classification after inspecting the red job state:

• Downgrade job 87304536524
• stable Core job 87308303933
• LTS Core job 87308303941

All three were on self-hosted-4vcpu-8gb-* runners and were marked failed while their test step remained in_progress, with cleanup/post-checkout still pending. They are runner losses, not test failures. The pre Core, both AD lanes, both Enzyme lanes, QA, all downstream jobs, Runic, and spelling checks pass. No failure is attributable to the one-line workflow trigger change.

[ChrisRackauckas-Claude]

Rebased the branch onto current main (a56b7cb) after verifying the fork head still matched 9a230e6 exactly. The focused Downgrade.yml main-branch trigger diff is unchanged. Runic passed locally, then I pushed empty co-authored retry commit e522369 to rerun the three jobs that previously lost their self-hosted runners; the retry commit has no tree changes.