fix(degree): bound DNF expansion so render_plan_graph can't hang

[lionelle]

Second phase of the MCP tester feedback fixes (independent of #12 — touches different files).

Issue 4 — render_plan_graph "Tool execution failed" on sample / plan_index≥2

shortest/longest rendered fine; every other selector failed with a generic error. Investigation showed:

• The picker (pick_by_category / plan_index) is correct and unit-tested, and render_plan_graph is panic-guarded (catch_unwind) — so a panic would surface as JSON, not "Tool execution failed".
• All the tester's calls shared one cache key (max_plans=None) → the analysis ran once and succeeded (shortest rendered) → the failure is in per-plan rendering, and the generic message is a client-side timeout on a hang.

Root cause: build_edges_from_courses calls parse_to_dnf on each in-plan course's prerequisite string, and parse_dnf_recursive expands an AND-of-ORs by cartesian product. A pathological (A|B) & (C|D) & … (~30 groups) is 2³⁰ paths and effectively hangs. That course only appears in the electives a sample/calc-ready plan draws — which is exactly why the extremes worked.

Fix

Cap the DNF expansion at MAX_DNF_PATHS = 1024 inside parse_dnf_recursive, bounding both the AND cartesian product and the OR accumulation. Every kept path is still complete (one course per AND group) and valid; we just stop enumerating further alternatives. This protects all three parse_to_dnf callers (render, plan_export, course_graph), not just the render path. No realistic prerequisite has near 1024 genuinely distinct satisfying paths, so existing behavior is unchanged (all prior parser tests still pass).

The plan's optional defense-in-depth render wall-clock guard was not needed: the DNF cap removes the only superlinear step, and the rest of the render is O(nodes+edges), bounded by plan size.

Verification

• New tests: a 2³⁰-path AND-of-ORs and a 5000-way OR return bounded + fast in 0.04s (would hang unbounded); a render_plan_graph of a plan whose course carries a pathological prerequisites_raw completes in 0.09s.
• cargo fmt --check, cargo clippy --all-targets --all-features -D warnings, cargo doc -D warnings, cargo test --all-features all green.

🤖 Generated with Claude Code