Fix dtype mismatch in NemotronH/Zamba2 Mamba2 CUDA-kernel path (out_proj)

[yuekaizhang]

Summary

The Mamba2 SSM CUDA-kernel forward path (cuda_kernels_forward) projects the scan output
through out_proj without casting it to the projection weight dtype. Because the SSM kernels
(mamba_chunk_scan_combined / selective_state_update) return fp32 regardless of the module
dtype, the projection fails whenever out_proj.weight is a narrower dtype than the activations:

RuntimeError: expected mat1 and mat2 to have the same dtype, but got: float != c10::BFloat16

Fix

Cast the projection input to the out_proj weight dtype before both explicit out_proj calls in
the CUDA-kernel path, so the matmul dtypes always agree (matching the slow path's intent, and
robust to mixed-dtype models):

out = self.out_proj(scan_output.to(self.out_proj.weight.dtype))

NemotronHMamba2Mixer inherits this method from Zamba2MambaMixer, so the fix is applied in the
modular source models/zamba2/modular_zamba2.py and propagated to the generated
modeling_zamba2.py and modeling_nemotron_h.py.

[github-actions]

[For maintainers] Suggested jobs to run (before merge)

run-slow: nemotron_h, zamba2

[Rocketknight1]

Yes, LGTM! Seems like a clean fix, and should be low risk because it's at worst a no-op.

[github-actions]

CI Dashboard: View test results in Grafana

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.