feat(#11): add context file validation script

[sugat009]

Summary

• Add npm run validate-context script that validates agent-memory/ context files against schema.json
• Align schema.json frontmatter definition with TEMPLATE.md fields (id, category, domain, issueNumber, issueUrl, title, lastUpdated, summary, services, techStack)
• Add 24 unit tests covering frontmatter parsing, field validation, body section checks, and file discovery

Context

Issue #11 had 4/5 acceptance criteria already met via #1, #20, and #72. This PR addresses the remaining two:

• Validation script works (the npm run validate-context command)
• JSON schema tested (unit tests for schema-based validation)

All 10 existing context files in agent-memory/domains/forms-and-reports/issues/ pass validation.

Test plan

• npm run validate-context passes all 10 existing context files
• npm run validate-context -- path/to/invalid-file.md reports errors for malformed files
• npm test passes all 154 tests (130 existing + 24 new)
• npm run lint shows no new errors

[Hareet]

Background

Based on the memory-distillation-pipeline, we need to converge our schema.json and template.md with the human-reviewed. There's a few good sections inside the distillation-pipeline that would be beneficial to us. I'll have a comment on that PR with some updated requests. I've ran the memory-distillation-pipeline for 1k PRs, and am trying to combine the format to open up per-domain bulk context PRs.

Review hiccups across 2 PRs that we could handle here:

Thanks for picking up the schema/TEMPLATE.md alignment here — a couple of things to reconcile before this lands, since it overlaps the provenance schema work in #109:

Date-field conflict with #109. This PR aligns schema.json to TEMPLATE.md using lastUpdated (camelCase) and requires it; #109 adds provenance fields using last_updated (snake_case). As written, whichever of the two merges second will leave one population of files failing validation. Could you reconcile the two here — e.g. accept the date under either name via an object-level anyOf ([{required:[last_updated]},{required:[lastUpdated]}]) and pick one as canonical going forward — so the schema is coherent once both land?

A few related divergences worth folding in while you're here: category appears as both improvement and enhancement across the corpus (treat as synonyms?), and the interoperability files use cht-interoperability-<n> ids, so an id pattern of ^cht-core-... would reject them — ^cht-[a-z0-9-]+-[0-9]+$ covers both.

Validation coverage. The "all existing context files pass" check looks like it ran against ~10 files; validating the full domains/ set surfaces a few real data issues the schema should catch: 4 files carry techStack values in services: (config, xml, xlsform, typescript) and use shared-libs (decide whether shared-libs joins CHTService or those get corrected), and 10729-smsparser-bugs-typos.md has an unquoted summary: with a colon that's invalid YAML. Might be worth running the new validate-context script across all of domains/ to confirm the full corpus is green.