Avoid intermediate QUBO files for MQLib runs

[bernalde]

Summary

• Stream the generated QUBO model to MQLib through /dev/stdin on Unix platforms, avoiding an intermediate .qubo file on disk.
• Keep the existing tempfile-backed path as a fallback for platforms without /dev/stdin, including Windows.
• Add focused tests for MQLib input selection and model serialization.
• Add David E. Bernal Neira to the package authors and README authors.

Note: the current MQLib_jll build recipe only installs ExecutableProduct("MQLib"); it does not provide a shared-library product or C ABI. This PR isolates the executable input boundary and removes the file write where the current artifact supports doing so, but it does not yet implement direct shared-library access.

Follow-up work for direct shared-library access

• Add a stable C ABI for solving QUBOs with MQLib #9: Add a stable C ABI for solving QUBOs with MQLib.
• Build MQLib_jll with a shared library product #7: Build MQLib_jll with a shared library product.
• Use the MQLib shared library from MQLib.jl #8: Use the MQLib shared library from MQLib.jl.

Tests run

• julia +1.12 --project=. test/runtests.jl - passed (MQLib input: 4 passed; QUBODrivers suite: 124 passed).
• julia +1.12 --project=. -e 'using Pkg; Pkg.test()' from /tmp/mqlib-ci-test.RMVVVu, a CI-like copy without the repo's ignored local Manifest.toml - passed.
• git -c core.whitespace=cr-at-eol diff --check - passed.

Notes about tests not run

• julia +1.12 --project=. -e 'using Pkg; Pkg.test()' directly in this checkout was not usable because the ignored local Manifest.toml pins path dependencies from this machine and conflicts with the current QUBOTools checkout version. The CI-like temporary-copy run above excludes that ignored manifest.
• Windows CI was not run locally.

Refs #3

[bernalde]

Question: This implementation still invokes the MQLib executable and streams the serialized model through stdin on Unix; it does not provide direct shared-library/C-API access. Since #3 specifically asks for direct access to MQLib as a shared library, should the PR body use Refs #3 instead of Closes #3 unless maintainers intend to accept this stdin path as the full resolution?

No code-level blocking issues found in the diff.

Tests run:

• julia +1.12 --project=. test/runtests.jl passed (MQLib input: 4 passed; QUBODrivers suite: 124 passed).
• julia +1.12 --project=. -e 'using Pkg; Pkg.test()' passed from /tmp/mqlib-review-test.KjtI3U, a clean copy without the ignored local Manifest.toml.
• Manual default-hyperheuristic smoke test through MQLib.Optimizer() passed.
• gh pr checks 6 reports Ubuntu and Windows CI passing.

[bernalde]

Closing this as superseded by the shared-library path in #15.

The useful part of this PR is the Unix-only /dev/stdin optimization for the executable fallback. That is no longer the primary path once MQLib_jll exports libmqlib_c_api, and this branch now overlaps the same sampler/test code that #15 rewrites. Merging it now would create avoidable conflicts without advancing #7 or #8.

If we still care about optimizing the executable fallback after #15 lands and MQLib_jll v0.1.2+0 is available, the right follow-up is to port only the fallback input-selection pieces into _sample_with_executable with focused tests. I would not carry over the unrelated author metadata changes as part of that follow-up.