docs: scrub private library references, update CI to public airframe dep

- AGENTS.md: rewrite for current architecture (airframe is public crates.io dep)
- .github/copilot-instructions.md: remove stale private path dep / AIRFRAME_ACCESS_TOKEN docs
- .github/workflows/ci.yml: remove AIRFRAME_ACCESS_TOKEN and private airframe clone steps
- .github/workflows/version-validation.yml: same
- .github/workflows/release-dry-run.yml: update stale comment about private path dep
- docs/MODEL_EXPANSION.md: libshimmy -> airframe (6 occurrences)

Author: Michael-A-Kuykendall

.github/copilot-instructions.md

@@ -5,47 +5,35 @@ Keep this file limited to current operating rules. Remove stale content aggressi
 ## Deployment Model (Read This First)
 
 ```
-shimmy-private/          ← THIS REPO — public-facing CLI/server product
-                            airframe = { path = "../", optional = true }
-../airframe/             ← PRIVATE sibling repo — GPU engine (NOT nested here)
+shimmy-private/              ← THIS REPO — public-facing CLI/server product (private working copy)
+public remote: shimmy.git    ← https://github.com/Michael-A-Kuykendall/shimmy.git
+airframe = { version = "0.1" }  ← PUBLIC crates.io dependency — no local path dep
 ```
 
-- **Shimmy is the product.** Airframe is the private optional engine dependency.
-- The `airframe` crate is OPTIONAL (`optional = true`). Default builds (`cargo build`) do not require it.
-- `cargo build --features airframe` requires `../airframe` to exist (private repo, not cloned by default).
-- **Cloning shimmy does NOT expose airframe source.** This is by design.
-- Do NOT add `airframe` as a nested submodule inside this repo. It is a sibling path dep only.
+- **Shimmy is the product.** Airframe is the GPU engine published on crates.io as `airframe = "0.1"`.
+- `airframe` is a crates.io dependency — no private path dep, no AIRFRAME_ACCESS_TOKEN, no cloning required.
+- `cargo build` (default, all features) works for anyone — airframe downloads from crates.io.
+- No submodules. No secrets needed for standard builds.
 
 ## Feature Flags
 
 ```toml
-default = ["huggingface"]        # crates.io-safe — no airframe required
-airframe = ["dep:airframe"]      # optional; requires ../airframe path dep (private)
-gpu = ["airframe", "huggingface"]
+default = ["airframe", "huggingface"]  # Full GPU build
+airframe = ["dep:airframe"]            # Airframe native GPU engine (crates.io: airframe = "0.1")
+gpu = ["airframe", "huggingface"]      # GPU-optimized build via Airframe
 full = ["airframe", "huggingface", "mlx"]
-fast / coverage = ["huggingface"]  # CI-safe, no path deps
+fast / coverage = ["huggingface"]      # CI-safe, no GPU hardware required
 # Deprecated stubs (llama.cpp removed in v2.0):
 llama = []  llama-cuda = []  llama-vulkan = []  llama-opencl = []
 ```
 
 ## Repository Push Policy
 
-- This repo's submodule checkout has **ONE** remote: `private` → `https://github.com/Michael-A-Kuykendall/shimmy-private.git`
-- **NO `origin` remote exists in the submodule context. Do not add one.**
-- The standalone checkout at `C:/Users/micha/repos/shimmy-private` has `origin` → same URL.
-- Push only when explicitly requested: `git push private <branch>` (submodule) or `git push origin <branch>` (standalone).
-- Do not push to public shimmy without explicit user approval.
-
-## CI Wire — AIRFRAME_ACCESS_TOKEN
-
-Confirmed wired in `.github/workflows/release.yml` and `ci.yml`:
-
-```yaml
-- name: Checkout Airframe engine (private path dep)
-  run: git clone https://x-access-token:${{ secrets.AIRFRAME_ACCESS_TOKEN }}@github.com/Michael-A-Kuykendall/airframe.git ../airframe
-```
-
-`AIRFRAME_ACCESS_TOKEN` must be set in GitHub Secrets. **Do not print or expose this token.**
+- This repo has two remotes:
+  - `origin` → `https://github.com/Michael-A-Kuykendall/shimmy-private.git` (private working repo)
+  - `public` → `https://github.com/Michael-A-Kuykendall/shimmy.git` (public GitHub repo)
+- Push to `public` when releasing or making docs/CI changes visible to users.
+- Push only when explicitly requested by the user.
 
 ## Test Failures
 
@@ -56,14 +44,14 @@ Do not declare work complete while any test is red.
 
 ## Architecture (v2.0)
 
-- Engine: wgpu/WebGPU WGSL pipeline via Airframe (private). Replaces llama.cpp entirely.
+- Engine: wgpu/WebGPU WGSL pipeline via Airframe (crates.io: `airframe = "0.1"`). Replaces llama.cpp entirely.
 - Server: OpenAI-compatible (`/v1/chat/completions`, `/v1/completions`), Ollama-compat (`/api/generate`, `/api/tags`).
 - No Python dependencies in the default build path.
 - WGSL quant coverage: F32, F16, Q4_0, Q8_0, Q4_K(M/S), Q5_K(M/S), Q6_K.
 
 ## What NOT To Do
 
-- Do NOT add an `airframe/` submodule inside this repo (was done by a previous AI session — removed).
-- Do NOT add an `origin` remote to the submodule checkout.
+- Do NOT add an `airframe/` submodule inside this repo.
+- Do NOT add a local path dep to airframe — use `airframe = { version = "0.1" }` from crates.io.
 - Do NOT mix console (`crates/console/`) or vision feature work into runtime release changes.
 - Do NOT push without explicit user request.