Expose multivariate normal method argument in post-estimation tasks

[jessegrabowski]

Post-estimation sampling tasks, especially sample_conditional_posterior, can be really slow for long time series. One reason for this is that I had set the method argument of all the multivariate normal distributions internally to be svd, which is the most robust option. There are cases, however, where one knows the model is well behaved, so you can drop back to something faster (like cholesky) without any problem. This PR allows this choice.

The default is still svd. Ideally we'd check a couple covariance matrices from the provided idata and try to give the user a smart default, but I don't want to do that much work here.

[ricardoV94]

do you want to type hint the options and/or mention them on the docstrings? Potentially mention the reason for choosing some over the other?

[jessegrabowski]

Just pushed this without having read this comment, so yes :)

[Copilot]

Pull Request Overview

This PR exposes the internal multivariate normal method argument (default "svd") in post-estimation sampling, forecasting, and related state‐space APIs, allowing users to choose faster but potentially less robust algorithms like "cholesky" or "eig".

• Added method: str = "svd" parameters to distribution constructors (__new__, dist, rv_op) in filters/distributions.py and to public sampling, forecasting, and IRF methods in core/statespace.py.
• Propagated the method argument through all internal pm.MvNormal.dist(..., method=...) calls.
• Updated docstrings in core/statespace.py to describe the new method parameter and its allowed values.

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

File	Description
pymc_extras/statespace/filters/distributions.py	Expose method argument in distribution constructors and propagate to MvNormal
pymc_extras/statespace/core/statespace.py	Add method parameter to sampling, forecasting, and IRF methods with docs

Comments suppressed due to low confidence (1)

pymc_extras/statespace/filters/distributions.py:213

• Add tests for multiple method values (e.g., 'cholesky', 'eig') in the pm.MvNormal calls inside the scan to ensure the new argument is correctly applied.

mu=0, cov=Q, rng=rng, method=method