Solve single-variable QUBO inputs exactly instead of crashing

[bernalde]

Summary

minimize/maximize crashed on single-variable (1x1) QUBO inputs because the underlying ITensor DMRG solver does not support system sizes of 1:

ERROR: `dmrg` currently does not support system sizes of 1.

This adds an exact single-variable path in _minimize: when the tensorized Hamiltonian has one site, it evaluates the objective at x = 0 and x = 1, returns the minimizer's energy (or reports degeneracy as a uniform superposition when the two are equal), and constructs a Solution consistent with the multi-variable path (energies/bond_dims/elapsed_times populated, iteration log and on_iteration callback honored). No public API or default behavior changes for multi-variable inputs.

This edge case currently has slightly divergent ad hoc fixes duplicated across open PRs #32, #33, #34, and #41. Landing it once here lets those PRs drop their copies and avoids merge conflicts.

Tests

Added four testsets to test/qubo.jl under "QUBO Correctness": single variable, single variable with linear and constant terms, single-variable degeneracy, and single-variable maximize. On unmodified main the first of these throws (dmrg ... system sizes of 1); with this change they pass.

Commands and results (Julia 1.10.11):

• Reproduction on main (clean re-resolved env): minimize(reshape([-2.0], 1, 1)) throws ErrorException: dmrg currently does not support system sizes of 1.
• julia --project=. -e 'using Pkg; Pkg.test()' on this branch: passed (Testing TenSolver tests passed), QUBO Correctness 82/82 (was 78 on main), all other testsets green (Ill-formed input, PUBO, Logging, JuMP, QUBODrivers.jl, Aqua.jl, VRP, HDF5, Doctests).

Notes

• Local Manifest.toml files are gitignored and must be re-resolved for the active Julia version; a stale 1.12-resolved manifest causes Pkg.test() to fail on 1.10. Removing it forces a fresh resolve, which is what CI does. No manifest is committed.
• No new dependencies; no CI/workflow changes.

Closes #49

[bernalde]

Blocking issues: none.

Nonblocking issues: none.

Questions: none.

Tests run and outcomes:

• Confirmed PR metadata and diff basis: merge-base ca2788038c2528a312a5315752d45ce2c0dff40a; local diff matched GitHub totals, 2 files changed, +71/-0.
• julia --project=. -e 'using Pkg; Pkg.test()' on the PR head: passed.
• julia --project=. -e 'using Pkg; Pkg.instantiate(); using TenSolver; minimize(reshape([-2.0],1,1); verbosity=0)' on origin/main: reproduced dmrg currently does not support system sizes of 1.
• PR-head smoke for single-variable callback/stat behavior: passed.
• PR-head smoke for a single-variable DynamicPolynomials input under the test project: passed.

Merge-readiness: I found no blocking issues. The change addresses #49, and the Closes #49 linkage matches the issue scope. This PR is still a draft, and because I am the PR author I am submitting this as COMMENT; formal approval must come from another maintainer.

[iagoleal]

We should now edit #32, #33, #34, and #41 to remove the redundancy