feat(ai): add optional backend commands

[CoderMungan]

No description provided.

[CoderMungan]

This PR is the full implementation of Block A: AI Backend as specified in
issue #92 (specs/ctx-ai-backend.md).

What ships:

• internal/backend/ — Backend interface, Registry, and OpenAI-compatible
  HTTP client
• internal/backend/vllm.go + openaicompat.go — vLLM and generic
  openai-compatible backends
• internal/backend/builtin_internal.go — thin wrappers for openai,
  anthropic, ollama, lmstudio
• internal/cli/ai/ — ctx ai ping and ctx ai extract commands (Option 1:
  new ctx ai top-level namespace)
• internal/cli/setup/core/backend/ — ctx setup --backend <name> extension;
  templates endpoint + auth wiring into .ctxrc
• internal/rc/ — .ctxrc backends: YAML table parsing and validation
• internal/compliance/ai_boundary_test.go — deterministic-core boundary guard;
  CI fails if ctx agent / ctx status / hooks ever import internal/backend
  (structural enforcement of Invariant 2)
• internal/cli/ai/core/run/ — validation consumer: schema-constrained
  completion → proposal artifact write (.context/*.md files are never touched
  directly)
• docs/cli/ai.md, docs/cli/setup.md, docs/recipes/local-inference-with-vllm.md
  — CLI reference and recipe docs

Related issues and PRs:

• Closes ctx ai backend #92
• Related: feat(backend): add llama.cpp (llama-server) backend wrapper #97 — llama.cpp (llama-server) backend wrapper (separate
  contributor, out of scope for this PR)