docs: v1.1 plan — multi-GPU + VynGraph gap closure

Comprehensive spec for the next major release focused on:

VynGraph integration gaps (P0/P1):
1. Actor migration between GPUs — quiesce/transfer/resume
2. Live actor introspection — streaming metrics via K2H
3. Hot rule reload — OWL/SHACL via HotReloadManager::RuleSet
4. Per-message provenance — PROV-O metadata in envelope
5. Multi-tenant K2K isolation — per-tenant sub-brokers

Multi-GPU core prerequisite:
- P2P mapped memory (CUDA IPC)
- NVLink topology detection (nvmlDeviceGetNvLinkState)
- Multi-GPU runtime facade
- Cross-GPU K2K messaging

Formal verification plan:
- TLA+ models for migration, ordering, tenant isolation
- Runtime assertion framework (feature-gated)
- 8-test hardware matrix on 2×H100
- Documentation of proven properties

Release plan: 4 phases over ~11 weeks, targeting 2× H100 Azure VMs
(ideally single-VM NVLink like ND96isr_H100_v5).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: mivertowski