Ship more built-in fitfuns (elastic-net, top-q-by-coefficient)

[gregfaletto]

Inspired by reviewing CRAN's stabs.

What stabs does

Ships several ready-to-use fitting functions so users don't have to write their own: glmnet.lasso, glmnet.lasso_maxCoef, lars.lasso, lars.stepwise — each conforming to a documented fitfun contract.

What cssr has today

cssr supports arbitrary fitfuns via css(..., fitfun = ...) and checkCssLoopOutput() validates their output, but it ships only one built-in: cssLasso (glmnet lasso at a fixed λ). Users wanting anything else must implement and validate their own.

Proposed scope

Add one or two ready-to-use fitfuns with the documented contract (returns an integer vector of selected feature indices), e.g.:

• an elastic-net variant (glmnet with alpha < 1), and/or
• a "top-q by |coefficient|" selector (select a fixed number of features, useful for the error-control q parameter in Cluster-level error control (PFER bounds): expose the paper's Shah-Samworth cluster guarantees (ss.thm.1/2) #87).

Each ships with tests and an example, mirroring how cssLasso is documented.

Notes

• Lower priority: cssr's niche is clustered features, so the marginal value is moderate.
• A "top-q" fitfun pairs naturally with the error-control work (Cluster-level error control (PFER bounds): expose the paper's Shah-Samworth cluster guarantees (ss.thm.1/2) #87), where q (features selected per subsample) is part of the cutoff/q/PFER triangle.
• Watch dependencies: an elastic-net fitfun reuses the existing glmnet dependency; avoid adding new hard deps (e.g. lars) — gate any such helper behind Suggests + a requireNamespace check if added.

References

• stabs: ?fitfun, ?glmnet.lasso, ?lars.lasso.
• cssr: cssLasso, checkCssLoopOutput.