Sync launchable notebooks with source updates and add Mean-Variance s…

[phuo-nv]

Summary

• Sync notebooks/launchable.ipynb and brev/launchable-content.ipynb with their source notebooks (cvar_basic, efficient_frontier, rebalancing_strategies) so updates introduced in Add Mean-Variance optimizer, refactor to shared base classes, and update notebooks #32 / Require python 3.11 in order to resolve conflicts with UV sync command #34 / fix notebook cuOpt presolve setting: False -> 0 for cuOpt 26.4 compat #36 / notebook API compat for cuOpt 26.4 + pydantic settings migration #38 propagate: new colored callout <div> styling on every section heading, regime_name rename, ApiSettings(...) pydantic model, presolve: 0, num_scen adjustments, etc.
• Add a new Mean-Variance Portfolio Optimization section (mean_variance_basic.ipynb content) to both launchable variants as ## 10. Mean-Variance Portfolio Optimization, including Mathematical Formulation, GPU solve via CVXPY + cuOpt Python API, Results, CPU comparison, and Summary table.
• Renumber Advanced topics continuously so Efficient Frontier = ## 8, Rebalancing = ## 9, Mean-Variance = ## 10 (was restarting at 1). Both ToCs updated.
• Heading-level cleanup: Strategy 1 / Strategy 2 are now ### under section 9 (were ##); Summary is now ### under section 10 with its "Mean-Variance vs CVaR" subheading bumped to ####.
• Disambiguate the duplicate ### Install Dependencies headings in notebooks/launchable.ipynb and brev/launchable-setup.ipynb: the first one (CUDA version detection) is now ### Detect CUDA Version, the second (the actual install) keeps ### Install Dependencies. ToC updated.
• Fix QA-test regression introduced in e65a857: compare_results() now displays the cuOpt GPU solver as cuOpt (GPU) regardless of whether it was invoked via CVXPY (str(cp.CUOPT) → "CUOPT") or the cuOpt Python API (already "cuOpt"). Other CVXPY backends are suffixed (CPU). The previous output printed raw CUOPT which broke the QA test pattern r"cuOpt.*?(\d+\.\d+)" for two cases:
  • [cvar_basic] Solve CVaR Optimization, Compare results between GPU and CPU solvers
  • [launchable_brev] Comprehensive validation - CVaR, Efficient Frontier, Rebalancing
• Drop redundant --- separators from Mean-Variance subsection cells (they came from source where each was top-level H2; here they're nested H3), the empty preamble cell, and a 4-newline gap inside the Mean-CVaR Introduction.
• Fix brev/launchable-content.ipynb ToC formatting: extra --- between items 9 and 10 removed; the kernel-spec note is now properly separated.
• Build cvxpy from upstream master instead of the PyPI wheel (via [tool.uv.sources] git source). The Mean-Variance Markowitz QP solve through cp.CUOPT relies on the set_quadratic_objective_matrix() wiring in cuopt_conif.py that's on master ahead of the next cvxpy release. Verified locally: cp.CUOPT solves the Markowitz QP and matches cp.CLARABEL to 1e-9 on objective and weights. uv.lock pins commit 766bdb92 (= cvxpy 1.9.0.dev0); bump with uv lock --upgrade-package cvxpy.

Test plan

• uv lock --locked passes (165 packages, no resolution drift)
• ruff format --check src/ tests/ passes (17 files already formatted)
• ruff check src/ tests/ passes
• pytest tests/ passes — 41 passed, 1 documented skip (was 38 before, +3 new tests in TestCompareResultsLabels that pin the QA pattern contract)
• Heading hierarchy in both launchable notebooks has 0 H-level skips (perfect H1 → H2 → H3 → H4 nesting); 0 empty markdown cells; 0 back-to-back --- separator cells
• Body cells of notebooks/launchable.ipynb and brev/launchable-content.ipynb are byte-identical (only the install/verify preamble differs, by design)
• Anchors (<a id="mean-variance"> etc.) reachable from both ToCs
• Post-merge: main.yml papermill run on GPU runner regenerates fresh outputs with cuOpt (GPU) labels — QA notebook tests for [cvar_basic] and [launchable_brev] should pass

Files changed

File	Change
src/utils.py	Add _solver_display_label() helper, use it in table rows and Objective Differences lines
tests/test_core.py	Add TestCompareResultsLabels (3 tests pinning the QA-pattern contract)
notebooks/launchable.ipynb	Rebuilt body from source notebooks; new MV section; renumbered Advanced topics; heading + line-break cleanup; disambiguated install headings
brev/launchable-content.ipynb	Same body as launchable (minus install/verify preamble); ToC + heading cleanup
brev/launchable-setup.ipynb	Disambiguated Install Dependencies → Detect CUDA Version for the first heading

Note on committed notebook outputs

The .ipynb files still embed outputs from the previous compare_results call (showing CUOPT). The QA tests parse the re-executed *_result.html produced by main.yml's papermill step, not the committed outputs, so the test will pass after merge. Re-executing locally to refresh the committed outputs would take ~30 min on the H200 — happy to do that as a follow-up if reviewers prefer fresh outputs in the diff.