Remove the shared compat entries from test and docs Project.toml

[michakraus]

Part R14 of the plan "Unify the test suites" (a removal part of its §8).

test/Project.toml and docs/Project.toml no longer carry a [compat] entry for a dependency of the root Project.toml:

• test/Project.toml: GeometricBase = "0.14", GeometricEquations = "0.21", Symbolics = "7" are removed.
• docs/Project.toml: Symbolics = "7" is removed.

Both environments contain the package through [sources], so the resolver applies the root's bounds to every shared dependency. An entry in test/ or docs/ can only duplicate or narrow them. The test-only and docs-only bounds do not change. Nothing under src/ changes. A CHANGELOG.md entry is added.

The test suite gives the same result on both trees: 218 pass on the branch and 218 on origin/main. The 9 testsets have the same counts, and the resolved versions are identical.

Pre-PR verification

• Fixed: nothing needed a fix.
• Unresolved: none. The stale comment in .github/dependabot.yml:8 is covered by the separate workflow rollout, and this PR does not edit .github/.
• Pre-existing: (nit) CHANGELOG.md:304-305 is an earlier [Unreleased] entry. It describes test/Project.toml with the root's bounds repeated. It stays true in effect, and the changelog rule does not allow a correction.
• Checked and clean: Each removed entry is the same as the root [compat] bound of that direct dependency. Both environments get the package through [sources] or Pkg.develop, so resolution does not change. The Julia floor is 1.11, so [sources] works on every CI version. Aqua test_deps_compat and the Downgrade job read only the root project. No manifest is tracked. Both TOML files parse. There is no stale prose outside CHANGELOG.md. The changed files are NFC-normalised.
• Not checked: piracy, inference and allocations (no executable change), the doctests and the docs build (the docs resolution does not change), and the test suite (CI runs it).

Critic

Verdict: PASS, round 1, at 1a1952b. Judged: origin/main...HEAD, with origin/main at d404c28.

clause	verdict
test-layout.jl --check gives no violation: exit 1 with 4 lines on the base, exit 0 with no output on the branch	PASS
Exactly the shared entries go: 4 lines are removed, and no shared key is left	PASS
No other line changes; test-only and docs-only entries stay	PASS
A comment that names a removed bound goes with it: no such comment exists	PASS
Resolution does not change: the test env and the docs env resolve identical versions on both trees	PASS
run-tests.jl full is green and no test is lost; CI is green	not verified by the critic; the dispatcher's full run gives 218 = 218, and CI on this PR shows the rest
A CHANGELOG entry names the removed entries and the rule	PASS
Mutant: add back Symbolics = "7" in docs/	CAUGHT
No change under src/; only the named files change	PASS

Other defects: 1. not verified: CI green on every job, the Documentation job included. CI on this PR answers it.

Known issues added

none

Advisor decisions

Decided (advisor, 2026-09-28, medium): every remaining J, A and R part. Where the build-part skill or julia-builder says "record the other defects" and ~/.claude/rules/known-issues.md says "a defect that the branch causes stays out", known-issues.md wins, as the M-part decision in §4 (line 523) already applied it. (1) Branch-caused: an other defect is branch-caused when it is absent on origin/main and a line that the branch adds or changes causes it. Examples are a wrong or false comment, docstring, CHANGELOG line or KNOWN_ISSUES.md line, a duplicate check, a new docstring on no docs page, and a test weaker than its own name. The builder fixes it in the same fix commit as that round's blocking defects, keeps the fix to what the defect needs, and records no entry for it. The next verify critic judges it as part of the fix diff. A critic fails a branch that records such a defect instead of fixing it. (2) After a PASS, a branch-caused other defect in text only (a comment, CHANGELOG.md, KNOWN_ISSUES.md, a docs Markdown file) is fixed, and the verifier runs once more, with no critic round and no suite run. A defect in code or in a test goes to the advisor. (3) Not branch-caused, and recorded with its kind: a pre-existing defect; an upstream fault in another repository (the J Revise entry, user decision in §7, is one); a limit of a form that the plan or the user decided; and a not verified doubt that no run of the part answers. Each such entry goes in the PR body under ## Known issues added. (4) A doubt that a run the part already requires will answer (a CI job of Done when, a full run) is no entry. The PR body names it, and the run answers it before merge (known-issues.md: "a doubt that a later run answers … its entry is deleted"). (5) A3, round 1: K17 is branch-caused. Add "integrators/solver_defaults.jl" to the Pages list under "Geometric Integrator" in docs/src/modules/integrators.md (a small stale premise of A3 Files), and delete K17. K21 is answered by A3 Done when 5 (the min job). Delete K21, and name it in the PR body. K18 stays (not verified): LU() is the user's decision (§8, option a), and no run of the part measures its cost. K19 stays (missing test): the walk is the guard that the §8 advisor decision chose, and it cannot see a future override that drops the default. Correct its evidence, because each of the four present overrides has a read-back. K20 stays (upstream). Replace its evidence, which cites a probe that is gone ("earlier version"), with a short snippet and its output on origin/main.

🤖 Generated with Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[codecov]

Welcome to Codecov 🎉

Once you merge this PR into your default branch, you're all set! Codecov will compare coverage reports and display results in all future pull requests.

Thanks for integrating Codecov - We've got you covered ☂️