docs: bootstrap docs/architecture/ with ADRs 0001 and 0002

[orwellsanimal]

Summary

Establishes Tier 2 of the architecture-grounding strategy: a docs/architecture/ directory with MADR-lite Architecture Decision Records as the durable record for non-obvious choices. Adds the bootstrap files (README, template) plus the first two ADRs, both of which backfill decisions already made and exercised in this repo.

What's an ADR, and why now?

We already had three places that captured something about decisions:

• CLAUDE.md — conventions and rules the agent must follow today. Mutable, present-tense.
• Memory (feedback_*.md, project_*.md) — observations, gotchas, current state. Point-in-time, decays.
• Commit messages / PR descriptions — the what of each change. Scattered across history.

None of those answers the question "why did we pick this fork in the road, and what did we reject?" a month or six months later. ADRs are short, dated, immutable markdown files that capture exactly that — the choice, the rejected alternatives, the consequences we accepted. They keep the workspace coherent across sessions (human or AI) and explain to future contributors how the repo got here.

Format is MADR-lite: Status / Context / Options considered / Decision / Consequences / Related. One page is the target; two is the ceiling. Once Accepted, an ADR is immutable — supersede with a new one, don't edit history.

What's in this PR

File	Purpose
docs/architecture/README.md	Index, when-to-write rules, status lifecycle, ADR table
docs/architecture/template.md	MADR-lite skeleton for new ADRs
docs/architecture/0001-update-set-as-artifact.md	Backfills 2026-05-15: why global-scope changes ship as deterministic update-set XML artifacts through the companion servicenow-coder-global repo, reviewed in a PR and imported via Studio + CICD API. Captures the four rejected alternatives (Studio-linked repo, PySNC direct writes, accepting the gap, ServiceNow DevOps) and the deliberate trade of speed for audit.
docs/architecture/0002-issue-pr-merge-workflow.md	Backfills 2026-05-19: why every change post-demo goes feature-branch → PR → squash-merge, but branch protection is deliberately off (no second human reviewer exists in a solo workspace, so required-reviews would either block work or train self-approval muscle-memory).

Both ADRs reflect choices that have already shipped and been exercised in the repo — 0001 is the basis for PRs #1, #2, #4, #6 in servicenow-coder-global; 0002 is the basis for every PR in this repo from #1 (2026-05-20) forward, including this one.

Tier context

The Tier 2 bootstrap is recorded in project_architecture_grounding.md (memory). Tier 1 (SDK + examples + platform-docs submodules) was already in place. Tier 3 (additional ServiceNow GitHub repos as submodules) and Tier 4 (CTA blueprint topics) remain deferred until a concrete need surfaces.

ADR backlog

Three pre-identified decisions remain on the backfill list, to be written one at a time as architecture comes up:

1. catalog_item_patch (partial-update artifact for the update-set generator)
2. Thin integration glue (external code forwards verbatim; mapping/lookup lives in SN tables)
3. Script Include bridge for scoped APIs (sn_sc, sn_ws) in module context

Test plan

• README.md ADR table renders correctly on GitHub and links resolve
• Both ADRs follow template.md structure (Status / Context / Options / Decision / Consequences / Related)
• ADR-0001 cross-refs match what's actually in the repo (scripts/build-update-set.js, scripts/python/preview-update-set.py, scripts/import-update-set.js, scratch/global-changes/)
• ADR-0002 cross-refs match real artifacts (PRs feat: hybrid import-update-set.js for global-scope changes #1, Track: GlideAggregate contribution to ServiceNow/PySNC #7, refactor(python): run aggregate tools on upstream PySNC GlideAggregate #9; memory file feedback_pr_workflow.md)
• Self-review the rendered diff once more on the PR before merging

🤖 Generated with Claude Code