Make solver adapter APIs preparation-aware

[termoshtt]

Depends on

• #1167, which adds the Rust and protobuf output-objective model used by this Python integration.

Summary

• Make SolverAdapter.solve() and SamplerAdapter.sample() non-mutating convenience APIs that copy the input, apply the adapter's recommended preparation policy for INPUT_CLASS, and delegate to new preparation-free solve_strict() / sample_strict() APIs.
• Expose the Instance-owned OutputObjective through the Python SDK, generated stubs, and API reference so transformed solver inputs still produce source-semantic objective values and sense.
• Update HiGHS, PySCIPOpt, Python-MIP, and OpenJij adapters to implement the strict API and preserve their adapter-specific options.
• Transport backend optimality only when the output objective records that the active-formulation proof remains valid; keep dual transport as a separate conservative rule.
• Keep Experiment solve and sampling records as the original input plus returned output, without storing the temporary prepared Instance.
• Update the English and Japanese migration guide, adapter tutorial, capability guide, special-constraint guide, OpenJij examples, and release notes.

User impact

Most callers can pass an unprepared Instance directly to solve() or sample() without having their object mutated. Applications that need custom preparation choices can prepare an explicit working copy and call the strict API. Custom Adapter implementations move their exact-input execution logic to solve_strict() or sample_strict().