Add constraint intelligence operating layer

[leadingproblemsolver]

Summary

Canonicalizes the full human-reviewed discovery + constraint-intelligence operating layer directly onto main, superseding the nested PR #1 dependency.

What changed

• preserves issue bodies and labels in discovery artifacts instead of throwing away downstream evidence
• uses production-constraint searches across runtime truth, side effects, identity/credentials, persistence/resume, delivery, resource pressure, configuration fidelity, and buyer/workaround cost
• adds a versioned root-constraint taxonomy
• adds build_constraint_intelligence.py to classify signals, build economic chains, score constraint value separately from intervention value, assign saturation states, and route actions
• adds compile_constraint_corpus.py to merge/deduplicate multiple signal sources before compilation
• warm-starts the corpus with explicitly tagged observed-summary seed signals
• keeps default public action as no comment; human review remains mandatory
• retains the GitHub issue discovery/review queue from the former parent PR
• fixes the live GitHub Search 422 by batching broad term families so each API query stays within GitHub's boolean-operator limit
• adds regression coverage for that query batching

Outputs

• artifacts/github-review-queue.json / .md
• artifacts/constraint-intelligence.json
• artifacts/constraint-leaderboard.json
• artifacts/constraint-intelligence.md

Evidence boundary

Seed records are observed summaries, not verbatim issue bodies. Constraint classification/scoring remains a heuristic operational-analysis layer; this does not expand LCE product claims.

Canonicalization rule

This PR now targets main directly. PR #1 is superseded by this complete dependency-correct head and should not be merged independently.

Verification gate

Core CI and the live scheduled discovery/compile workflow must both pass on the repaired head before merge.