Fix verbose handling in OptimizationMOI

[ChrisRackauckas-Claude]

Summary

Fixes a crash when verbose is passed to a MOI optimizer through OptimizationMOI. There are two manifestations, both from the same root cause — __map_optimizer_args was never updated for the OptimizationVerbosity system, while verbose flows into the cache's solver_args:

1. FieldError: type MOIOptimizationNLPCache has no field verbose — the @SciMLMessage calls read cache.verbose, but the MOI caches (MOIOptimizationNLPCache / MOIOptimizationCache) keep solve kwargs in solver_args and have no verbose field. Triggered whenever a common tolerance (abstol/reltol/maxiters) is passed alongside verbose.
2. Unable to add option "verbose" ... must be ::String, ::Integer, or ::Float64 — verbose was splatted from solver_args into the raw optimizer-attribute loop, so even a plain verbose=-solve (no tolerance) failed by trying to set verbose as an Ipopt RawOptimizerAttribute.

Fix

__map_optimizer_args now consumes verbose as a named keyword (so it is no longer forwarded into the raw-attribute loop) and runs it through OptimizationBase._process_verbose_param to obtain a valid OptimizationVerbosity for the @SciMLMessage calls. The now-unused cache argument is dropped and both call sites updated.

How it surfaced

Reported via the BoundaryValueDiffEq optimization-based BVP path (the rocket optimal-control example), which builds an OptimizationProblem and calls solve(prob, Ipopt.Optimizer(); abstol=..., verbose=OptimizationVerbosity()) — hitting manifestation (1).

Testing

• New @testset "verbose kwarg" covering all three cases (verbose-only, verbose+abstol, verbose=true). Each reproduces a pre-fix failure and passes after.
• Full OptimizationMOI Core suite run locally on Julia 1.12.6 — all green (the 2 Broken in cache are pre-existing @test_broken):

  verbose kwarg | 3  3  3.1s
  

• End-to-end: the original rocket BVP script (MIRK4(; optimize = Ipopt.Optimizer())) now returns retcode = Success (max height 1.0128323), matching the OptimizationIpopt result.

---

This PR is a draft and should be ignored until reviewed by @ChrisRackauckas.

🤖 Generated with Claude Code

https://claude.ai/code/session_01RtJAi7bALH5XkakjS6Yb6j

[ChrisRackauckas-Claude]

Pushed two follow-up commits addressing the OptimizationMOI QA (Aqua) failures, which are pre-existing on master (confirmed by running QA against a clean master worktree — same failures, independent of the verbose fix):

• Allowed piracy: the SciMLBase solver-trait methods defined on the MathOptInterface optimizer types are an intentional interface extension, now declared via Aqua.test_all(...; piracies = (treat_as_own = [MOI.AbstractOptimizer, MOI.OptimizerWithAttributes],)). Aqua now passes 11/11.
• deps_compat: added the missing [compat] bound ReverseDiff = "1" for the test-only [extras] dependency.

The third QA finding (a JET error on MOIOptimizationNLPCache's setproperty! setfield! fallback) is unrelated to this PR and is fixed separately in #1236. I verified locally that with this PR and #1236 applied, JET.test_package passes cleanly (0 errors) and Aqua passes 11/11 — the two together turn the QA job green.

Should be ignored until reviewed by @ChrisRackauckas.