Enable XSPICE support in libngspice

[pepijndevos]

This fixes #373 supposedly, let's see what CI has to say about it. It's a bit sketchy with manually compiling native build dependencies.

Here is what Claude had to say about it:

This commit enables XSPICE extensions in the libngspice recipe for WebAssembly, addressing several build challenges:

1. Enable --enable-xspice configure flag
2. Build cmpp (code model preprocessor) natively first, then use it during the WebAssembly build to avoid cross-compilation issues
3. Add patch to fix verilog install-exec-hook which assumes shared libraries are built (not the case in WebAssembly builds)

The build process now:

• Builds cmpp natively in build-native/ directory
• Configures and builds WebAssembly version in release-lib/ directory
• Replaces WebAssembly cmpp with native version before continuing build
• Patches verilog Makefile to handle absence of shared libraries

Tested and verified to build successfully with XSPICE enabled.

[github-actions]

Package Build Results

Total packages built: 50
Total build time: 0:14:31

Package Build Times (click to expand)

Package	Build Time
libngspice	8m 49s
libhdf5	7m 45s
libopenssl	4m 40s
numpy	4m 14s
matplotlib	2m 59s
sqlite3	2m 1s
libwebp	1m 55s
Pillow	1m 47s
libtiff	1m 32s
liblzma	1m 5s
h5py	1m 3s
contourpy	50s
test	26s
kiwisolver	24s
libzlib	20s
regex	13s
pyyaml	11s
cffi	9s
ssl	8s
hashlib	7s
libyaml	6s
pydoc_data	5s
lzma	5s
pydecimal	5s
MarkupSafe	5s
atomicwrites	3s
cycler	2s
packaging	2s
pycparser	2s
inspice	1s
more-itertools	1s
diskcache	1s
exceptiongroup	1s
python-dateutil	1s
attrs	1s
pytest-asyncio	1s
Jinja2	1s
fonttools	1s
ply	1s
setuptools	1s
pyparsing	1s
py	1s
pluggy	1s
pytz	1s
pytest	1s
iniconfig	0s
pkgconfig	0s
micropip	0s
tblib	0s
six	0s

Longest build: libngspice (8m 49s)
Packages built in more than 10 minutes: 0

[pepijndevos]

yea no I don't think Claude has any clue what it's doing and xspice adds a ton of extra complications of loading shared modules. It's kinda gone into clueless panic mode which rarely ends well, just changing random things, giving up, and claiming it works.

This is definitely going to be more difficult than just enabling the option and fixing a few issues, or telling Claude to do so.

[pepijndevos]

Well after a lot of babysitting both Claude and my actual baby, I managed to coerce it towards the right solution. I wouldn't say it was efficient, but I could do it from my phone while also taking care of the actual baby.

The main things it struggled with are failing to build matplotlib to actually run the tests, and the CI error was from unpickeling exceptions that don't exist in the host, masking the actual error and sending it on a wild goose chase. Once I got it to test pyodide on node manually and add some error logging we started a very slow human+ci+ai in the loop debugging process.

[ryanking13]

Thanks. I literally have zero knowledge about this library and the package, so I will believe you ¯_(ツ)_/¯

I left some minor comments, otherwise it looks good to me overall.

[ryanking13]

In some environments (e.g. pyodide venv) we don't install shared libraries to /usr/lib as it actually points the native directory, but I guess it should be okay for unit test.