fix(zamba): add use_associative_scan config flag to avoid torch.compile slowdown

[msnliu]

ZambaMambaMixer.forward() passes use_associative_scan=False directly to mamba_selective_scan, which falls back to a sequential Python loop that makes torch.compile slow.

The other Mamba-family models (mamba, jamba, falcon_mamba) instead read this setting from their config and forward it to the scan function. This PR adds the same plumbing to ZambaConfig and ZambaMambaMixer.

Changes:

1. configuration_zamba.py: add use_associative_scan: bool = True field
2. modeling_zamba.py: read config.use_associative_scan in init and pass it in forward() instead of the hardcoded False

@vasqu

[github-actions]

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

run-slow: zamba

[github-actions]

---

CI recap

Dashboard: View test results in Grafana
Latest run: 32953204500
Result: failure | Grafana metrics are not available yet.

Code quality check failed: test jobs were skipped. Fix the code quality issues and push again to run tests.