test(observability): cover console-metrics, batch, reqlog & readiness branches

[mauripunzueta]

Stacked on top of #150 (feat/observability-uptime). Adds tests for the genuinely-uncovered behavioral branches that PR #150's Codecov report flagged.

Context

Most of PR #150's reported patch-coverage gap is not new-test-addressable:

• error-handling closures (.map_err(|e| internal_error(...)) on SQL failures) — unreachable without fault injection;
• stale carryforward — code that already has in-tree tests but shows uncovered because Codecov's snapshot predates this branch's own test commits (a fresh coverage run clears these);
• Docker/OTLP-bound — mongodb/postgres/composite/telemetry.

This PR covers the branches that were actually uncovered and cheaply testable.

Tests added

• reqlog — 3xx / 4xx / sub-200 status-class classification arms (existing tests only exercised 2xx/5xx).
• console resource-distribution — the other roll-up bucket taken when distinct types exceed top.
• console tenants — a tenant present only in request-log traffic (reported with resources: 0).
• batch — exhaustive TransactionError -> (status, code) mapping and the status_text reason-phrase table.
• health — the /_readiness handler response shape.

Adds helios-observability as a dev-dependency of helios-rest so the console integration test can seed the in-process request log directly (the test router has no request-tracking middleware).

CI note

Because the repo gates pull_request CI to base: main, this stacked PR (base feat/observability-uptime) will not run CI/Codecov until #150 merges and GitHub auto-retargets it to main.