feat: log aggregation and conditional alert rules

[btcneves]

Objective

Adds two independent features to AutoFlowOps:

1. Log aggregation — structured log shipping to Loki and Elasticsearch with a consistent label schema across all service components.
2. Conditional alert rules — per-job rules that trigger Alerts based on configurable conditions evaluated after every execution.

---

Changes

Log Aggregation

• Structured JSON logging via structlog with context propagation through contextvars
• Loki and Elasticsearch log shipping; configurable via environment variables (LOG_SINK, LOKI_URL, ELASTICSEARCH_URL)
• Consistent label schema: service, env, job_id, execution_id, workspace_id
• docker-compose.observability.yml with Loki + Promtail stack for local development
• Grafana provisioning datasource for Loki auto-configured
• docs/log-aggregation.md: label schema, sink configuration, local setup and production guidance
• Deployment docs updated with observability environment variables

Conditional Alert Rules

• AlertRule model with CASCADE FK to jobs, workspace isolation via SET NULL FK to workspaces
• Alembic migration c3d4e5f6a7b8 creates the alert_rules table with index on job_id
• Four condition types: http_status_gte, duration_ms_gte, response_body_contains, consecutive_failures_gte
• CRUD API under GET|POST /api/jobs/{job_id}/alert-rules and PATCH|DELETE /api/jobs/{job_id}/alert-rules/{rule_id}
• RBAC: viewers read-only; operators and admins manage rules (enforced via require_operator)
• evaluate_alert_rules() in http_runner.py; Celery worker wired post-retry to avoid premature alerts during retrying state
• React section in JobDetailPage with create form, inline toggle and delete; operators-only write actions
• Client-side validation mirrors backend constraints (integer thresholds, HTTP status range 100–599)
• i18n in English and Portuguese-Brazilian
• Audit logging on all mutations via log_action()
• API reference documentation updated with all four endpoints and curl examples

---

Test Plan

• uv run pytest — 260 backend tests pass (7 new alert rule tests, 3 new notification tests)
• npm test in frontend/ — 76 frontend tests pass (1 new JobDetailPage test)
• npx tsc --noEmit in frontend/ — no TypeScript errors
• uv run ruff check backend/ — no linting errors
• Alembic migration applies cleanly on a fresh database: alembic upgrade head
• Log aggregation: set LOG_SINK=loki and LOKI_URL=http://localhost:3100 — verify structured logs reach Loki
• Alert rule CRUD via API: create a rule, trigger an execution, verify Alert is created
• Workspace isolation: confirm rules are not accessible across workspace headers

---

Notes

• Log aggregation is additive and backwards-compatible; the default sink is stdout (existing behavior)
• Alert rules are evaluated after every execution regardless of execution status, enabling both failure-based and performance-based alerting
• consecutive_failures_gte queries the last 50 completed executions to determine the streak
• Frontend bundle includes a pre-existing chunk size warning (803 kB) unrelated to this PR

---

Security Checklist

• No real secrets, tokens, passwords or API keys in any committed file
• .env.production is not tracked
• All credentials masked in API responses, UI and delivery records
• No sensitive configuration details exposed in public documentation
• No git push --force, no history rewrite, no --no-verify bypass