Bump stale Optim/StaticArrays/Mooncake compat floors

[ChrisRackauckas-Claude]

Note

This PR should be ignored until reviewed by @ChrisRackauckas.

What

Three [compat] lower bounds predate the ForwardDiff 1 / Symbolics 7 modernization already reflected elsewhere in [compat], leaving the declared floors mutually unsatisfiable:

Dep	Was	Now	Why
Optim	1.7.5, 2	1.12, 2	Optim 1.7.5 requires ForwardDiff 0.10, conflicting with the ForwardDiff = "1" floor. Optim 1.12 is the first release that supports ForwardDiff 1.
StaticArrays	1.9	1.9.7	1.9 (= 1.9.0) is too old for DifferentiationInterface 0.7, which Symbolics = "7.0.0" pulls in. DI 0.7.2+ requires StaticArrays = "1.9.7 - 1".
Mooncake	0.4, 0.5	0.4.175, 0.5	Mooncake 0.4.0 requires DifferentiationInterface ≤ 0.6.16, conflicting with the Symbolics = "7.0.0" floor (needs DI ≥ 0.7.12). The lowest Mooncake supporting that DI range is 0.4.175.

Verification (local, Pkg's resolver)

• Floors resolve: pinned to the new minimums, the full test-dependency set resolves on Julia 1.10.8 → Optim 1.12.0, StaticArrays 1.9.7, Mooncake 0.4.175, Symbolics 7.0.0, DifferentiationInterface 0.7.15, ForwardDiff 1.0.0, NLSolversBase 7.10.0.
• Ceiling unaffected: the normal test environment still resolves → Optim 2.1.0, StaticArrays 1.9.18, Mooncake 0.5.31, Symbolics 7.26.0, DifferentiationInterface 0.7.18, ForwardDiff 1.4.1.
• Before this change, Pkg.resolve at the old floors fails with Optim 1.7.5 vs ForwardDiff 1, then StaticArrays 1.9.0 vs DI 0.7, then Mooncake 0.4.0 vs Symbolics 7 — all independent of the Julia version.

Scope / caveat

This PR fixes the genuinely-stale floors. It does not, on its own, turn the Downgrade CI job green. That job's remaining failure is driven by two separate issues, both outside this change:

1. The downgrade action resolves at the Julia major.minor floor (1.10.0), and modern Mooncake (needed for the Symbolics 7 / DI 0.7 floors) requires Julia ≥ 1.10.8 — so the merged test-dep graph is unsatisfiable at 1.10.0 regardless of these floors (julia = "1.10" is intentionally kept here).
2. The action's resolver (Resolver.jl, via julia-downgrade-compat Add Akima interpolation method #49's merged-extras path) reports Unsatisfiable even on graphs that Pkg.resolve resolves successfully (verified on the identical project file at Julia 1.10.8).

Those are upstream/tooling concerns to be handled separately; this PR is the correct-compat-floors piece and is independently valid.

🤖 Generated with Claude Code