Refactor path search around Pareto fronts

[isPANN]

Why

The previous path contract collapsed named ProblemSize components into a scalar total, treated symbolic analysis failure as an orderable growth result, and exposed implicit single-route winners through cost functions, top-level path output, and automatic reduction routing. That could discard incomparable measured vectors, recommend an unanalyzable route, and execute an arbitrary representative from a Pareto front.

Contract replacement

• Symbolic search excludes Growth::Unknown routes from the Pareto front while returning each excluded route with its fields and analysis-failure reason.
• Search completeness and symbolic analysis coverage are reported separately. Complete all-Unknown searches return NoAnalyzablePath; incomplete bounded searches remain explicitly incomplete.
• Measured search compares terminal ProblemSize vectors componentwise, removes dominated vectors, preserves incomparable vectors, and deterministically deduplicates only equal vectors by hops and stable path order.
• ProblemSize::total(), scalar cost/ranking APIs, and sum-based default budgets are deleted.
• Measured budgets are named per-field limits validated against registered size fields and checked after each constructed intermediate problem is measured. A legal field absent on one node is not fabricated.
• CLI, MCP, examples, callers, and docs now consume Pareto fronts explicitly. reduce requires one explicit continuous route; there is no top-level recommendation or automatic target routing.

Verification

• Added an independent fixed-seed oracle over 2-7 node graphs: simple-path enumeration plus an independent terminal componentwise filter is compared with production Pareto search.
• Added negative controls for incomparable, dominated, and equal measured vectors; Unknown exclusion and all-Unknown failure; known, unknown, and node-missing budget fields; post-construction budget enforcement; incomplete Unknown-only observations; and discontinuous explicit routes.
• CARGO_BUILD_JOBS=1 cargo test -p problemreductions --lib rules::graph::pareto_tests:: -- --test-threads=1 (32 passed)
• CARGO_BUILD_JOBS=1 cargo test -p problemreductions --lib rules::graph::tests:: -- --test-threads=1 (61 passed)
• CARGO_BUILD_JOBS=1 cargo test -p problemreductions --lib test_reduction_graph:: -- --test-threads=1 (49 passed)
• Focused CLI/MCP front, reduce, solve-bundle, extract, route-continuity, and inspect tests passed single-threaded.
• Core clippy passed with -D warnings; CLI clippy passed with the existing tool_router dead-code warning.
• cargo fmt --all -- --check and git diff --check passed.

Tests were intentionally focused; no heavyweight workspace coverage or large path enumeration was run. Builds used CARGO_BUILD_JOBS=1 and tests used one thread.

Explicitly not implemented

This PR does not provide a scalar recommendation, weights/priorities/constraint-based route selection, a compatibility or deprecated winner API, memory/byte estimates, worker isolation, allocation metadata, or new resource fields. Callers must choose a Pareto-front route explicitly.

This is a stacked PR based on 1075-growth-domain; it must not be merged independently of that base.