Ship juliapkg.json with the documented solver stack (fixes #175)

`load_julia_packages` was running `Pkg.activate("diffeqpy", shared=true)`
before adding the solver packages, which moved Julia off juliacall's
managed project (`$CONDA_PREFIX/julia_env`, where PythonCall is pinned
to the ABI-matched version) and into a separate shared environment. The
shared environment then resolved an unconstrained PythonCall, breaking
the juliacall <-> PythonCall ABI and segfaulting on import.

Declare the solver stack in `diffeqpy/juliapkg.json` so juliacall
installs DifferentialEquations, OrdinaryDiffEq, StochasticDiffEq,
DelayDiffEq, Sundials, DiffEqCallbacks, and ModelingToolkit into its
managed project alongside PythonCall, with proper version resolution.
Compat ranges span the doc-tested major and the current major (e.g.
DifferentialEquations "7, 8") so the docs keep working under the
recent v8 metapackage where some sub-solvers are no longer transitive.

Drop the `Pkg.activate` line from `load_julia_packages` so the lazy
install path used by the GPU sub-modules (cuda/amdgpu/metal/oneapi)
also targets juliacall's managed project.

https://claude.ai/code/session_01G6GjTpy5ZKWq78zfdCzV4y

Author: claude

MANIFEST.in

@@ -2,3 +2,4 @@ include README.md
 include LICENSE.md
 include docs/*.txt
 include diffeqpy/*.jl
+include diffeqpy/*.json

diffeqpy/__init__.py

@@ -29,12 +29,19 @@ def load_julia_packages(*names):
     """
     Load Julia packages and return references to them, automatically installing julia and
     the packages as necessary.
+
+    Packages declared in ``diffeqpy/juliapkg.json`` (the documented solver stack:
+    DifferentialEquations, OrdinaryDiffEq, StochasticDiffEq, DelayDiffEq, Sundials,
+    DiffEqCallbacks, ModelingToolkit) are installed by juliacall at import time into
+    its managed project alongside the ABI-matched PythonCall, so they should already
+    be importable here. Anything not listed there (e.g. the optional GPU backends
+    pulled in by ``diffeqpy.cuda`` / ``amdgpu`` / ``metal`` / ``oneapi``) is added
+    lazily on first use into the same managed project.
     """
     # This is terrifying to many people. However, it seems SciML takes pragmatic approach.
     _ensure_julia_installed()
 
     script = """import Pkg
-    Pkg.activate(\"diffeqpy\", shared=true)
     try
         import {0}
     catch e

diffeqpy/juliapkg.json

@@ -0,0 +1,33 @@
+{
+    "julia": "1.10",
+    "packages": {
+        "DifferentialEquations": {
+            "uuid": "0c46a032-eb83-5123-abaf-570d42b7fbaa",
+            "version": "7, 8"
+        },
+        "OrdinaryDiffEq": {
+            "uuid": "1dea7af3-3e70-54e6-95c3-0bf5283fa5ed",
+            "version": "6, 7"
+        },
+        "StochasticDiffEq": {
+            "uuid": "789caeaf-c7a9-5a7d-9973-96adeb23e2a0",
+            "version": "6, 7"
+        },
+        "DelayDiffEq": {
+            "uuid": "bcd4f6db-9728-5f36-b5f7-82caef46ccdb",
+            "version": "5, 6"
+        },
+        "Sundials": {
+            "uuid": "c3572dad-4567-51f8-b174-8c6c989267f4",
+            "version": "4, 5, 6"
+        },
+        "DiffEqCallbacks": {
+            "uuid": "459566f4-90b8-5000-8ac3-15dfb0a30def",
+            "version": "3, 4"
+        },
+        "ModelingToolkit": {
+            "uuid": "961ee093-0014-501f-94e3-6117800e7a78",
+            "version": "9, 10, 11"
+        }
+    }
+}