Add spikeforge to the framework support table

[w4ffl35]

Adds spikeforge (BSD-3-Clause, pip install spikeforge) to the framework support table. It's an SNN toolkit built on snnTorch/PyTorch that reads and writes NIR through spikeforge.nir_bridge.

One thing I'd like your judgement on, rather than quietly ticking the box.

I've marked Read from NIR as ✓, but spikeforge's read path differs from the other rows and you may prefer ⬚. The other frameworks import a NIR graph into their own native model. spikeforge loads the graph and executes it directly through an independent interpreter — it does not reconstruct a native module:

from spikeforge.nir_bridge.ingest import load_external, interpret_file

graph = load_external("model.nir")       # NIRGraph, from any framework
result = interpret_file("model.nir", spikes)   # executed, never touching snnTorch

So a NIR file written by Norse or snnTorch runs in spikeforge and produces outputs, which I think is what the column means to a user — but it is genuinely not "convert to native". Happy to change it to ⬚ if you read the column the other way. Any node kind outside the shared vocabulary raises a typed error naming the node rather than being skipped.

Write is the conventional direction: to_nir(spec, module) emits a graph, verified against the 20-node conv_net preset.

Why the row might be worth having beyond one more entry: spikeforge re-executes every exported graph through that same independent interpreter and reports per-tensor numerical drift against the source model, so an export can be validated rather than assumed. As far as I can tell none of the listed frameworks does this, and it's a NIR-specific capability rather than a general feature:

pip install "spikeforge[nir]"
spikeforge-verify validate --topology conv_net

That exits non-zero outside tolerance. On conv_net today the worst deviation is 1.4e-07 mean-abs on the lif1 membrane.

Examples column: every other row points at a neuroir.org docs page. spikeforge has none, so I've linked the runnable example in-repo (examples/04_nir_export_validate.py, verified end to end on a clean install). If you'd like a proper docs page in the NIR docs tree instead, I'm glad to write one in a follow-up PR — I didn't want to add pages to your docs uninvited in a first contribution.

Docs: https://docs.spikeforge.net/interpreter-spine

[Jegp]

Thanks for the addition. I appreciate your question about the interpretation of reading/writing, but we're very loose here: if you can load a graph and run if, I'd tick the box, no matter how you do it. Same for writing.

I see you're using snnTorch behind the scenes, but what I think sets spikeforge apart is that you could use another simulator one day, the API isn't strictly tied to a simulator.

I'm happy to approve and merge as-is. My only suggestion would be to add an example in the example pages so it's clear how your framework is used. Basically, you can just copy in the code you wrote in this PR :-)