Add executed subsea-to-shore design and response-surface Colab notebook

[EvenSol]

Adds an executed, source-pinned Colab study following the response-surface method described in equinor/neqsim#4253. It compares PipeBeggsAndBrills and TwoFluidPipe and includes a fixed-temperature homogeneous AdiabaticPipe diagnostic. All design cases are synthetic; the public ASME abstract does not supply the paper's numerical dataset, so this is not a reproduction of its results or experimental model qualification.

The notebook solves hydraulic rates for a 100-to-80 bara receiving-pressure constraint across diameter, horizontal length and methane/decane composition. It retains all 15 executed code cells, 86 accepted rate roots, complete tables, three plots, surrogate coefficients, analytical and mesh checks, runtime provenance, and CSV/JSON export code. There are 27 training designs and 16 off-grid held-out designs per multiphase model. Every final root is replayed and checked; invalid TwoFluidPipe candidates never enter training.

At the 40 km / 0.35 m / 6 mol% decane baseline, rates are 40.38 kg/s (Beggs–Brill) and 44.04 kg/s (TwoFluidPipe). Model disagreement reaches 16.15% across the cases. A physics-informed quadratic correction reduces worst held-out surrogate error from 11.65% to 0.52% for Beggs–Brill and from 15.04% to 2.17% for TwoFluidPipe. Out-of-domain requests fall back to NeqSim. These errors are against each simulator, not against measurements.

Validation

• Exact current NeqSim master: 1768fc1fec78ff974fe9ff4232582f09ed504686, source-built 3.23.0 JAR with SHA-256 and class-location assertions.
• Three clean top-to-bottom executions through scripts/execute_notebook_inprocess.py; final notebook has sequential counts, retained outputs, zero stored errors and zero stderr.
• Ten engineering checks pass: mass balance, receiving-pressure residual, single-liquid Darcy/Colebrook limit, baseline mesh refinement, positive predictions, fallback and controlled defect reproduction. Maximum pressure-root residual is below 0.000079 bar; 30-to-60-section rate change is 0.096% for Beggs–Brill and 0.985% for TwoFluidPipe.
• Maven -DskipTests -Dmaven.javadoc.skip=true package: exit 0. Maven -Dtest=TwoFluidPipeSteadyStateConvergenceTest,PipeBeggsAndBrillsBoundaryTest test: six tests, zero failures/errors.
• Repository checker unit tests: six passed. Targeted scripts/check_notebook.py --require-main-source, using the focused new ledger and validation catalog: zero errors/warnings. NeqSim skill notebook checker, notebook schema, AST, <=100-character code lines and compact Colab math checks pass.
• All 329 relative catalog targets verified against the connector-read repository tree plus the new notebook.
• Rendered with nbconvert classic HTML and MathJax 3.2.2 SVG; every Markdown/result cell, all 16 math expressions, all tables and three figures inspected. WeasyPrint 70.0 provided a 13-page pagination view for visual QA.
• Repository-wide notebook integrity CI PASSED on this exact head: run 37612530896. Other inherited notebook-validation workflows are still running. No Java source changes are included in this PR.

NeqSim improvements

Verified and filed equinor/neqsim#4258: the native Beggs–Brill flow-rate solver returns normally on iteration exhaustion with a +0.43722 bar pressure residual versus its 0.008 bar tolerance. The notebook reproduces that defect separately and excludes it from all accepted design/training results.

Existing equinor/neqsim#2907, #4228 and #3298 are linked for public pressure-drop/holdup benchmarks, common network evidence and thermal/terrain/transient qualification. The public-case adapter requires source/reuse/units/uncertainty evidence and explicitly supports only the demonstrated binary, horizontal, isothermal setup. No duplicate campaign issues were created.

Documentation impact

Adds the notebook to README and the main examples catalog, introduces a dedicated maintenance-ledger shard and increments the active count from 321 to 322. Assumptions, installation/source provenance, evidence boundaries, output interpretation and improvement acceptance criteria are documented in the notebook.

Hosted-Colab downloads/source compilation have not been exercised locally; clean validation uses the exact source-built JAR through the documented local overrides. Experimental data, reservoir/well deliverability, heat loss, shore rise, equipment limits and transient operability remain outside this bounded hydraulic study.

Refs equinor/neqsim#4253. The research issue remains open because published-data validation is still outstanding.