[mandli]

This disables the testing on macOS to only every night to reduce action costs. Also updates versions of some actions and puts some additional guardrails on the CI.

---

Brings this repo's CI in line with GeoClaw (clawpack/geoclaw b432880d), where macOS was moved off the per-PR path because macOS runners bill at 10x the Linux minute rate.

Per-PR (testing.yml)

• Matrix is Linux only: ubuntu-latest, gcc 14, debug + opt. The gcc 15 toolchain and the os/toolchain exclude: block are removed.
• concurrency with cancel-in-progress so new pushes to a PR cancel the superseded run.
• paths-ignore for **.md, **.rst, docs/**, .gitignore (same list as GeoClaw).
• timeout-minutes: 30 (recent runs take 3–8 min).

Nightly (nightly.yml, new)

• schedule (07:30 UTC, staggered after GeoClaw's 06:00) + workflow_dispatch.
• macos-latest, gcc 15, debug + opt — the same macOS cells that previously ran per-PR, with the same test selection.
• No Linux leg: this repo has no slow- or remote-marked tests, so the per-PR Linux run already covers everything that Linux would.
• flake8 and Coveralls stay per-PR (Linux); the nightly runs plain pytest with the same -k "not test_shallow_sphere" filter.

Action versions (aligned with GeoClaw)

• actions/checkout@v7.0.1, actions/setup-python@v7.0.0, fortran-lang/setup-fortran@v1.10.0, actions/upload-artifact@v7
• test-summary/action@v2.6 (was v2.4).

Caveats

• Nightly failures do not block merges. For macOS-sensitive changes, run nightly.yml on the branch via Run workflow before merging.
• GitHub disables scheduled workflows after 60 days with no repo activity; it must be re-enabled from the Actions tab if that happens.

Verification

• This PR's checks show only ubuntu-latest jobs, all passing.
• After merge: dispatch nightly.yml on master and confirm both macOS legs pass.

[coveralls]

coverage: 47.07%. remained the same — mandli:ci-macos-nightly into clawpack:master