feat(tooling): runtime metric regression suites and JUnit reports

[meaningforger]

Summary

• Continue RFC-0090 after feat: add offline project compile regression suites #14 with metis project test --mode runtime, calling the production query_metrics service against deployment-owned placement and backends.
• Require declared external/live fixtures and complete schema/row expectations. Compare exact numeric strings without float conversion, preserve duplicate multiplicity, distinguish null, and support deterministic keyed tolerances and ordered assertions.
• Tighten private execution ceilings without mutating or relaxing deployment policy. Incomplete results, limits, timeouts, and execution outages cannot pass as expected semantic failures.
• Add private atomic JSON/JUnit output, bounded value-free diagnostics, fixture provenance, signal cancellation, and conflicting-output checks.
• Update README, shipped contract, RFC status, and independently reviewed Doris/ClickHouse fixture examples.

Validation

• make check
• make fmt-check
• CGO_ENABLED=1 go test -tags duckdb ./cmd/metis ./app/tooling/regression ./app/bootstrap -count=1
• Default tests cover strict schemas, exact numbers, duplicate/null semantics, tolerances and key uniqueness, report privacy, output safety, and private execution ceilings.
• Production DuckDB CLI integration covers success, mismatch, expected/unexpected semantic errors, and byte-limit failure. Doris/ClickHouse examples are bootstrap/schema-validated offline; no external warehouse fixture was provisioned during local verification.

Boundaries

This is the standalone local-unrestricted metric-result slice only. Comparison/attribution suites and host-managed policy assertions remain proposed; no fixture DDL, baseline approval, raw query override, or benchmark changes are included.