docs: align README + ai-skills + RELEASING with atomr formatting

* README rewritten to mirror atomr's structure: "Why ... in Rust, now"
  rationale, "What's in the box" crate table, separate "Quick start
  (Rust)" and "Quick start (Python)" sections, "Building from source",
  "Layout" tree, "Learn more" doc index.
* All "0.0" / "0.2.x" version references in docs/, ai-skills/,
  RELEASING.md, README.md updated to "0.1" (matches the v0.1.0
  rebrand cut on crates.io / PyPI).
* "atomr 0.2 integration / adoption" phrasing trimmed since the
  pinned-version flavor is no longer meaningful — atomr-accel just
  tracks atomr now.

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

Author: cognect

README.md

@@ -88,8 +88,8 @@ TestPyPI for wheel uploads.
 
 ```bash
 # crates.io
-cargo yank --vers 0.0.2 atomr-accel
-cargo yank --vers 0.0.2 atomr-accel-patterns
+cargo yank --vers 0.1.0 atomr-accel
+cargo yank --vers 0.1.0 atomr-accel-patterns
 # ...etc
 
 # PyPI: open the project page → Manage → Releases → Yank.

RELEASING.md

@@ -130,7 +130,7 @@ exists for code that genuinely benefits from it.
 
 | Backend | Status | Notes |
 |---|---|---|
-| **CUDA (NVIDIA)** | **Shipping** in `atomr-accel-cuda` | Full F2-F9 + atomr 0.2 integration. |
+| **CUDA (NVIDIA)** | **Shipping** in `atomr-accel-cuda` | Full F2-F9 + atomr integration. |
 | ROCm (AMD) | Designed for | hipBLAS / hipFFT / rocSPARSE / rocSOLVER map cleanly. RCCL ≈ NCCL. |
 | Metal (Apple) | Designed for | MPS + Metal command queues fit `AccelStream` directly. No NVRTC equivalent (use Metal compute shaders compiled at build time). |
 | oneAPI (Intel) | Designed for | SYCL + oneMKL libraries. |

ai-skills/skills/atomr-accel-backends/SKILL.md

@@ -50,7 +50,7 @@ pulls of the other blueprints.
 
 ```toml
 [dependencies]
-atomr-accel-cuda = "0.2"
+atomr-accel-cuda = "0.1"
 tokio = { version = "1", features = ["rt-multi-thread", "macros"] }
 ```
 
@@ -72,7 +72,7 @@ no-op when it's missing.
 | TelemetryExtension probes | `atomr-accel-cuda = { features = ["telemetry"] }` |
 
 Aggregates compose: `core-libs` ⊂ `training-libs` ⊂ `full-cuda`.
-The four atomr-0.2 integration features (`replay`, `cluster`,
+The four atomr integration features (`replay`, `cluster`,
 `streams`, `telemetry`) are independent of the library aggregates.
 
 For a goal-by-goal picker with explicit transitive-deps tables, see
@@ -82,8 +82,8 @@ For a goal-by-goal picker with explicit transitive-deps tables, see
 
 ```toml
 [dependencies]
-atomr-accel-cuda     = "0.2"
-atomr-accel-patterns = "0.2"  # adds DynamicBatchingServer, etc.
+atomr-accel-cuda     = "0.1"
+atomr-accel-patterns = "0.1"  # adds DynamicBatchingServer, etc.
 ```
 
 Only adds the patterns crate; `train` / `agents` / `realtime`
@@ -173,7 +173,7 @@ plumbing" checks.
   `DeviceActor::pre_start` spawns child actors and uses
   `tokio::spawn` internally — the runtime needs a multi-threaded
   scheduler.
-- **Pinning to `atomr-accel = "0.2"` with no `features = ["cuda"]`.**
+- **Pinning to `atomr-accel = "0.1"` with no `features = ["cuda"]`.**
   The umbrella crate ships only the trait surface by default; you
   also need `cuda` to get the concrete `DeviceActor`. Most projects
   should depend on `atomr-accel-cuda` directly instead.

ai-skills/skills/atomr-accel-getting-started/SKILL.md

@@ -71,7 +71,7 @@ panic!("ContextPoisoned: cublasCreate returned {status:?}");
 The application **never writes its own decider** unless it's
 running its own non-CUDA actor next to the device. For
 that, [`atomr-accel-cuda::error::DeviceSupervisor`](../../../crates/atomr-accel-cuda/src/error.rs)
-implements atomr 0.2's typed `SupervisorOf<ContextActor>` trait
+implements atomr's typed `SupervisorOf<ContextActor>` trait
 over `GpuError`, so you can pattern-match on the typed enum
 directly:
 

ai-skills/skills/atomr-accel-supervision/SKILL.md

@@ -36,15 +36,15 @@ are gated:
 | `observability::install`, `GpuProbes` | `telemetry` |
 
 ```toml
-atomr-accel-cuda = { version = "0.2", features = ["cudnn", "nvrtc", "replay"] }
+atomr-accel-cuda = { version = "0.1", features = ["cudnn", "nvrtc", "replay"] }
 ```
 
 For the by-goal picker with transitive dep impact, see
 [`docs/features-matrix.md`](../../../docs/features-matrix.md).
 
 ### "trait `Actor` is not implemented"
 
-You're using atomr 0.2 — the trait lives in `atomr_core::actor`.
+You're using atomr — the trait lives in `atomr_core::actor`.
 The cleanest import is the atomr-accel-cuda prelude, which
 re-exports the relevant traits:
 

ai-skills/skills/atomr-accel-troubleshooting/SKILL.md

@@ -330,7 +330,7 @@ crates/
 
 ## Status & roadmap
 
-`F2 – F9 implemented + atomr 0.2 adoption complete.` The project's
+`F2 – F9 implemented + atomr adoption complete.` The project's
 internal phase plan is "F"-numbered (foundation phases F1 through
 F10):
 

docs/architecture.md

@@ -38,7 +38,7 @@ workspace**. Reach for one without inheriting the others.
 
 ```toml
 [dependencies]
-atomr-accel = "0.0"
+atomr-accel = "0.1"
 ```
 
 Runtime cost: atomr-core + atomr-config + atomr-macros + cudarc +
@@ -48,8 +48,8 @@ tokio. cuBLAS is the always-on library; everything else is gated.
 
 ```toml
 [dependencies]
-atomr-accel          = "0.0"
-atomr-accel-patterns = "0.0"
+atomr-accel          = "0.1"
+atomr-accel-patterns = "0.1"
 ```
 
 Adds: nothing beyond the patterns crate itself.
@@ -58,7 +58,7 @@ Adds: nothing beyond the patterns crate itself.
 
 ```toml
 [dependencies]
-atomr-accel = { version = "0.0", features = ["nvrtc"] }
+atomr-accel = { version = "0.1", features = ["nvrtc"] }
 ```
 
 Adds: cudarc nvrtc bindings (zero new transitive deps; just unlocks
@@ -68,8 +68,8 @@ Adds: cudarc nvrtc bindings (zero new transitive deps; just unlocks
 
 ```toml
 [dependencies]
-atomr-accel       = { version = "0.0", features = ["full-cuda"] }
-atomr-accel-train = "0.0"
+atomr-accel       = { version = "0.1", features = ["full-cuda"] }
+atomr-accel-train = "0.1"
 ```
 
 `full-cuda` aggregate bundles cuDNN + cuFFT + cuRAND + cuSPARSE +
@@ -79,7 +79,7 @@ cuSOLVER + cuBLASLt + NVRTC + cuTENSOR + NCCL.
 
 ```toml
 [dependencies]
-atomr-accel = { version = "0.0", features = ["replay"] }
+atomr-accel = { version = "0.1", features = ["replay"] }
 ```
 
 Adds: atomr-persistence (Journal trait), serde, serde_json. The
@@ -91,7 +91,7 @@ on where you want events to land.
 
 ```toml
 [dependencies]
-atomr-accel = { version = "0.0", features = ["telemetry"] }
+atomr-accel = { version = "0.1", features = ["telemetry"] }
 ```
 
 Adds: atomr-telemetry. Run `atomr-dashboard` as a sidecar to
@@ -101,11 +101,11 @@ visualize.
 
 ```toml
 [dependencies]
-atomr-accel          = { version = "0.0", features = ["full-cuda", "replay", "cluster", "streams", "telemetry"] }
-atomr-accel-patterns = "0.0"
-atomr-accel-train    = "0.0"
-atomr-accel-agents   = "0.0"
-atomr-accel-cuda-realtime = { version = "0.0", features = ["nvrtc"] }
+atomr-accel          = { version = "0.1", features = ["full-cuda", "replay", "cluster", "streams", "telemetry"] }
+atomr-accel-patterns = "0.1"
+atomr-accel-train    = "0.1"
+atomr-accel-agents   = "0.1"
+atomr-accel-cuda-realtime = { version = "0.1", features = ["nvrtc"] }
 ```
 
 ## Feature reference

docs/features-matrix.md

@@ -12,7 +12,7 @@ A ten-minute tour. By the end you'll have:
 
 - Rust **1.78+** (workspace toolchain in `rust-toolchain.toml`).
 - A sibling clone of [atomr](../../atomr) at `../atomr` (path
-  dependency, version 0.2.x).
+  dependency, version 0.1.x).
 - Optional but recommended: an NVIDIA GPU plus the
   [CUDA Toolkit](https://developer.nvidia.com/cuda-toolkit) (driver
   + libraries). Without these, the workspace still builds and unit