docs(plans): add Plans 2-7 — full v1 implementation roadmap

Six plans completing the v1 implementation roadmap, written in parallel
by subagents (Plans 2, 3, 7 re-run with chunked Write/Edit strategy after
their original agents hit the 64k output-token cap).

- Plan 2 (245K, 35 tasks, 7104 lines): Full HermesInstance reconciler +
  validating/defaulting webhooks; all resource builders except
  hermes-runtime; HermesClusterDefaults singleton.
- Plan 3 (159K, 25 tasks, 4328 lines): Hermes-specific runtime --
  agent container image build, Python/uv init containers, multi-platform
  gateways (Telegram/Discord/Slack/WhatsApp/Signal), Honcho profile-store
  companion.
- Plan 4 (129K, 23 tasks, 3597 lines): HermesSelfConfig CRD + SSA-based
  reconciler with `hermes.agent/selfconfig` field manager; GitOps
  coexistence test.
- Plan 5 (235K, 32 tasks, 6215 lines): Day-2 ops -- S3 backup/restore,
  OCI auto-update with rollback, declarative OpenClaw migration; lesson
  #437 finalizer-via-r.Patch baked in.
- Plan 6 (176K, 24 tasks, 5249 lines): Distribution -- OLM bundle,
  release-please + GoReleaser, Cosign+SBOM, OperatorHub auto-submit;
  conformance suite (negative, idempotency, upgrade-matrix, GitOps,
  failure injection); k8s 1.28-1.32 matrix; perf benchmarks.
- Plan 7 (159K, 24 tasks, 4244 lines): v1 stability docs --
  api-versioning policy, deprecations policy, condition catalogue,
  ROADMAP, README polish, 9 worked examples, Grafana dashboard JSON,
  v1.0.0 release cut (with gated repo-public flip).

Known follow-up: Plan 3 uses spec-authoritative `botTokenSecretRef` /
`apiKeySecretRef`; Plan 6's conformance YAMLs may use a `tokenSecretRef`
shorthand. Alignment is implementation-time work, not blocking the plans.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: stubbi