docs: architecture and reading guide

[R7rainz]

What

Adds ARCHITECTURE.md — a guided reading path through the codebase, and links it from the README.

Understanding PulseOps currently means reverse-engineering why there's a Node backend and a Go prober talking over Kafka, before anything else clicks. This writes that reasoning down.

Contents

• The mental model — the one loop the whole system runs, and the single decision (probing is slow and network-bound; recording is transactional) that explains most of the architecture
• A twelve-stop reading path in dependency order, starting at schema.prisma and ending at docker-compose.yml, with a note on why each stop comes where it does
• The check pipeline as a mermaid diagram, plus the reasoning behind each concurrency guard — leader lock, SKIP LOCKED claim, manual Kafka commits, Redis dedup
• The data model grouped into five clusters rather than listed as 20 models, including why MonitorCheck is BigInt and composite-indexed
• Auth — the four credential paths converging on one revocable session, and the authn/authz distinction that the workspace-access middleware turns on
• Frontend, CLI/TUI/MCP, feature map
• Ideas worth stealing — the transferable patterns (SQL-table job queue, idempotency techniques, validate-at-use, fail-fast, liveness vs readiness)
• Exercises ordered by how much of the system they force you through
• Known gaps — stated plainly, including the absence of tests

Also

Corrects two stale README lines: the backend hasn't used node-cron since the Kafka scheduler landed, and the Go engine now probes TCP, DNS and keyword as well as HTTP.

Verification

Every file path referenced in the guide was checked to exist (34 paths + the dashboard route). Docs only — no code changes.