v0.2: Land Lemonade ADRs + migration plan + research docs

[thinmintdev]

What to build

Land the Lemonade migration design documents on main as a single review-and-merge PR. This is the review gate before any code work begins on v0.2.

Files to merge:

• docs/internal/adr/0006-migrate-inference-to-lemonade.md (strategic ADR)
• docs/internal/adr/0007-nuclear-evict-all-mitigation.md (tactical ADR)
• docs/internal/lemonade-migration-plan.md (full plan)
• docs/internal/lemonade-spike-findings-2026-05-22.md (spike data)
• docs/internal/lemonade-spike-runbook.md (spike runbook, reproducible)
• docs/internal/lemonade-repo-deep-dive-2026-05-22.md (research findings)
• CONTEXT.md — adds device term + revises slot term for v0.2 dual sense
• New memory pointers in /home/halo/.claude/projects/-home-halo/memory/MEMORY.md

Review focus: are the locked-in decisions reasonable given the spike data and research findings? Does the v0.2 scope make sense alongside the parallel Agents v0.2 work?

This is HITL — requires human review/approval before downstream issues begin.

Acceptance criteria

• All listed docs committed to main
• ADR-0006 status is Accepted (or revised + accepted)
• ADR-0007 status is Accepted (or revised + accepted)
• CONTEXT.md device and slot (v0.2 sense) terms reviewed
• Memory pointers indexed in MEMORY.md
• No remaining TODO or ??? markers in the merged docs

Blocked by

None — can start immediately

[thinmintdev]

In-flight via existing PRs

This issue's scope is already split across two merged/open PRs:

• ✅ PR docs: Lemonade migration strategy — ADRs + spike + runbook + slot definition #134 (merged, eb19ee1) — initial ADRs + spike findings + runbook + slot definition in CONTEXT.md
• 🔍 PR docs(lemonade): repo deep-dive + ADR/plan tightenings from research #138 (open) — research-aligned tightenings: ADR-0006 §3 HTTP-only + /v1/load schema, §5/§6 embeddable tarball + bare systemd (drops the custom container image), migration-plan matching, + lemonade-repo-deep-dive-2026-05-22.md (276L research artifact)

This issue closes when PR #138 merges. Review focus: are the post-research tightenings aligned with reality?

[thinmintdev]

Closing — both implementing PRs merged:

• PR docs: Lemonade migration strategy — ADRs + spike + runbook + slot definition #134 (eb19ee1, merged earlier) — initial ADRs + spike findings + runbook + slot definition
• PR docs(lemonade): repo deep-dive + ADR/plan tightenings from research #138 (f9df077, just merged) — research-aligned tightenings: HTTP-only, embeddable tarball + bare systemd, /v1/load schema documented, + lemonade-repo-deep-dive-2026-05-22.md

ADR-0006 + ADR-0007 on main reflect the post-research locked-in design.