Give pred path --all real Big-O output: delete the O(<raw expr>) fallback and JSON text rendering

[isPANN]

Background

Issue #1069: pred path --all used to OOM/hang because Big-O rendering fully expanded composed overhead expressions. A stopgap now caps expansion, but the CLI's failure path (problemreductions-cli/src/commands/graph.rs:349) falls back to printing the raw unreduced expression disguised as O(...), and path_all builds the text rendering unconditionally even in --json mode (both named in #1069). With big_o_normal_form now backed by the growth domain (#1078), every polynomial/linear-exponential path has a real answer, so the fallback and the wasted rendering can go — and #1069 can close with regression tests that pin the fix.

Full design (shared overview for this batch): docs/design/symbolic-growth-domain.md, sections M2/M4 and Quality requirements.

Objective

Delete the O(<raw expr>) fallback and the unconditional text rendering in JSON mode; add the regression, budget, and determinism tests that make #1069's failure mode structurally impossible to reintroduce. Closes #1069.

Interface (Input → Output)

• pred path <S> <T> --all [--max-paths N] (text): every listed path shows either a genuinely normalized O(...) per size field or an explicit annotation (e.g. not normalized: <reason>) — never a multi-hundred-character raw expression.
• pred path ... --all --json: emits JSON without constructing the text rendering.
• New tests: (1) a regression test that reconstructs pred path --all OOMs/hangs: big_o_normal_form exponentially expands composed overhead expressions #1069's exploding path (KSatisfiability → Satisfiability → KSatisfiability → DecisionMinimumVertexCover → HamiltonianCircuit → QuadraticAssignment → ILP → QUBO) via compose_path_overhead and asserts big_o_normal_form returns a genuine normal form; (2) a whole-graph budget test rendering Big-O for all paths returned by find_paths_up_to on representative hot pairs; (3) a golden-file determinism test for pred path KSat QUBO --all text output.

Technical recommendations (non-binding)

• Build the exploding path in the regression test by name (compose the edges via the graph API) rather than hard-coding the 2113-char expression, so the test tracks inventory changes.
• Keep each test < 5 s (repo policy).
• The golden test normalizes nothing: byte-exact comparison is the point (growth-term ordering is deterministic by construction from Add growth domain (src/growth.rs): growth terms, antichain, symbolic dominance #1075).

Verification

1. time pred path KSat QUBO --all (a release or debug build — state which in the test) completes in < 1 s and its output contains no line longer than 200 characters — which proves no raw-expression fallback remains.
2. cargo test issue_1069 runs the three new tests green.
3. Behavioral JSON check: pred path KSat QUBO --all --json | python3 -m json.tool > /dev/null exits 0, and the JSON contains a Big-O entry for every path element.

Negative control: the golden determinism test, when fed an output with two terms deliberately swapped (fixture included in the test), fails — proving the byte-exact comparison has teeth. Additionally, grep -n 'format!("O({})"' problemreductions-cli/src/commands/graph.rs returns nothing and the behavioral checks above still pass — proving the fallback was removed, not relocated.

Dependencies

Depends on #1078. Milestone: Symbolic Growth Domain & Pareto Search. Closes #1069.