feat(cuqdyn-c): always produce both covariance varieties

[ZocoLini]

The hidden states took their covariance one of two ways, picked with <uq_method>: the rank-aware FIM of CUQDyn1_Plus, or the hybrid that keeps the FIM marginal scale and takes its correlation from the leave-one-out ensemble. Choosing meant knowing in advance which one to trust, and comparing them meant a second full run.

Both are now built on every run. The expensive parts -- the leave-one-out ensemble, the forward sensitivities, the conformal bands -- are shared, so the second variety is one more quadratic form on a matrix that already exists. It does not cost enough to be worth a setting.

The result carries a UqBands per variety instead of one loose pair plus the q_low_alt/q_up_alt that used to appear only when the hybrid was selected, and the output file writes Q_low_<v>, Q_up_<v>, CovP_<v> and StdY_<v> for v in fim and hybrid. There is no bare Q_low/Q_up left: with no variety to select, no pair deserves the unqualified name. plot.py reads both shapes, so the CUQDyn1 reference dumps under example-results/ still plot, and it draws the hybrid band dashed over the FIM one on hidden states alone -- elsewhere the two coincide, since a conformal band has no covariance in it.

uq_method is gone from the XML schema, from CuqdynConf and from its Rust mirror, and the per-example cuqdyn-fim.xml / cuqdyn-hybridcov.xml pair collapses into one cuqdyn.xml.