Record verified solution values for five models

[bernalde]

Follows the solution-verification sweep in #152. No model code changes; READMEs only.

What changed

• cstr: the recorded value was stale. The README carried 3.0620145766; GAMS/BARON at optcr=optca=0 proves 3.062003537511728 with lower bound equal to upper bound. Updated, and labelled proven optimal.
• ex1_linan_2023, grid, multiperiod_blending, pandemic: these recorded no solution value at all. Each now carries the best value found under a stated budget (GAMS/BARON on gdp.bigm, 120 s, optcr=1e-6), explicitly labelled not proven optimal.

Every recorded point is backed by a GAMS/Examiner audit: primal and dual variable bounds and constraints at 1e-6, and both complementary-slackness conditions at 1e-7, with the discrete variables fixed. Each entry states its solver, tolerance, date, and audit status, so a later reader can tell a proven value from a best-found one and re-check either.

Two notes for reviewers:

• grid's optimum is 0.0, which is meaningful rather than degenerate: the objective minimizes total capacity slack, and zero slack turns out to be attainable for the default instance, so no expansion is required.
• multiperiod_blending ships 60 instances under instances_json/; the recorded value is for the instance built by the default build_model(), and the README now says so.

Models with PRs in flight are deliberately untouched here; their values should be recorded as those PRs land. water_network and reverse_electrodialysis are also untouched, since their recorded values need the discussions on #151 and #149 respectively (scaling, and a units/value discrepancy) before they can be stated cleanly.

Verification

pixi run pytest tests/ — 335 passed, 1 skipped.

Refs #152, #105