v0.4.1: fix notebook install — Python cell + install() function

v0.4.0's notebook install recipe used `%sh python3 genie-install`,
which loses the notebook kernel context. The Databricks SDK's
runtime-auth provider partially detects "looks like a notebook"
(Spark Py4J connects fine in the subprocess) but then fails the
IPython context lookup with `'NoneType' object has no attribute
'parent_header'` — and the SDK breaks out of the auth chain instead
of falling through to env vars. Result: install fails on the very
first turnkey notebook recipe people try.

scripts/genie-install: refactor `main()` to delegate to a shared
`_run(args, client)` helper and expose a new public entry point:

    def install(*, shared=False, main=False, version=None,
                bundle=None, target=None, repo=DEFAULT_REPO,
                keep_temp=False, client=None) -> int: ...

Callable from any Python context. CLI semantics unchanged.

docs/install/databricks-genie.md: TL;DR now leads with the Python-
cell + runpy pattern:

    %pip install --quiet databricks-sdk PyYAML
    import urllib.request, runpy
    urllib.request.urlretrieve(
        "https://github.com/easel/helix/releases/latest/download/genie-install",
        "/tmp/genie_install.py",
    )
    g = runpy.run_path("/tmp/genie_install.py")
    g["install"]()

Plus an explicit warning callout about why `%sh` doesn't work.

Tested via runpy.run_path + install() against /Users/.../helix-fn-test:
229/229 uploaded, 14/14 verify checks pass.

Author: easel

.claude-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "helix",
-  "version": "0.4.0",
+  "version": "0.4.1",
   "description": "HELIX methodology, artifact catalog, and routing skill for AI-assisted development.",
   "author": {
     "name": "Erik LaBianca",

docs/install/databricks-genie.md

@@ -8,20 +8,33 @@ design, review — over a project's governing artifacts.
 
 ### From inside a Databricks notebook (recommended — no setup)
 
-Paste this into a notebook cell. The Databricks runtime supplies
-implicit workspace credentials; no PAT, env vars, or CLI required.
+Paste this into a Python notebook cell. The kernel has implicit
+workspace credentials; no PAT, env vars, or CLI required.
 
 ```python
-%sh
-curl -fsSL https://github.com/easel/helix/releases/latest/download/genie-install -o /tmp/genie-install
-python3 /tmp/genie-install                  # user-scoped install
-# python3 /tmp/genie-install --shared       # workspace-wide (admin)
-# python3 /tmp/genie-install --main         # track main branch
+%pip install --quiet databricks-sdk PyYAML
+
+import urllib.request, runpy
+urllib.request.urlretrieve(
+    "https://github.com/easel/helix/releases/latest/download/genie-install",
+    "/tmp/genie_install.py",
+)
+g = runpy.run_path("/tmp/genie_install.py")
+
+g["install"]()                # user-scoped (current notebook user)
+# g["install"](shared=True)   # workspace-wide (admin)
+# g["install"](main=True)     # track main branch
 ```
 
-The installer auto-detects the workspace from the notebook's runtime
-context and uses your notebook identity for auth. Re-run any time to
-refresh.
+The `install()` function runs in the notebook's Python kernel where
+the Databricks SDK can use implicit notebook-runtime auth. Re-run any
+time to refresh.
+
+> **Don't use `%sh`** for the install. A `%sh` subprocess loses the
+> notebook's kernel context — the SDK partially detects "looks like a
+> notebook" (Spark Py4J connects) but then fails the IPython context
+> lookup with `'NoneType' object has no attribute 'parent_header'`.
+> Run the installer in a Python cell instead.
 
 ### From a dev box or CI
 

scripts/genie-install

@@ -438,14 +438,15 @@ def resolve_target(args: argparse.Namespace, client) -> str:
 # --- main -----------------------------------------------------------------
 
 
-def main() -> int:
-    args = parse_args()
-
+def _run(args: argparse.Namespace, client=None) -> int:
+    """Shared install body. Used by both the CLI entry point (`main`) and
+    the programmatic entry point (`install`)."""
     print(f"HELIX Genie installer")
     print()
 
-    info("connecting to workspace")
-    client = make_client()
+    if client is None:
+        info("connecting to workspace")
+        client = make_client()
 
     target = resolve_target(args, client)
     print(f"target:  {target}")
@@ -481,5 +482,51 @@ def main() -> int:
     return 0
 
 
+def install(
+    *,
+    shared: bool = False,
+    main: bool = False,
+    version: str | None = None,
+    bundle: "Path | str | None" = None,
+    target: str | None = None,
+    repo: str = DEFAULT_REPO,
+    keep_temp: bool = False,
+    client=None,
+) -> int:
+    """Programmatic entry point. Same semantics as the CLI.
+
+    Designed to be invoked from a Databricks notebook cell where the
+    kernel has implicit workspace credentials:
+
+        import urllib.request, runpy
+        urllib.request.urlretrieve(
+            "https://github.com/easel/helix/releases/latest/download/genie-install",
+            "/tmp/genie_install.py",
+        )
+        g = runpy.run_path("/tmp/genie_install.py")
+        g["install"]()                # user-scoped, latest release
+        # g["install"](shared=True)   # workspace-wide
+        # g["install"](main=True)     # track main branch
+    """
+    args = argparse.Namespace(
+        shared=shared,
+        main=main,
+        version=version,
+        bundle=Path(bundle) if bundle else None,
+        target=target,
+        repo=repo,
+        keep_temp=keep_temp,
+    )
+    if sum([bool(args.main), bool(args.version), bool(args.bundle)]) > 1:
+        fail(2, "--main, --version, --bundle are mutually exclusive")
+    if args.shared and args.target:
+        fail(2, "--shared and --target are mutually exclusive")
+    return _run(args, client=client)
+
+
+def main() -> int:
+    return _run(parse_args())
+
+
 if __name__ == "__main__":
     sys.exit(main())