[bump minor] Change to the observation covariance matrix to work with NaCl outputs

[corentinravoux]

NaCl outputs: covariance of 3N x 3N with full covariance for mb, x1 and c parameters (need to convert from x_0 to m_b)

[Copilot]

Pull request overview

This PR updates the SN Ia velocity data vector to correctly handle NaCl-style observation covariances that are 3N×3N over the SALT2 parameters (mb, x1, c), instead of an N×N distance-modulus covariance. It introduces a design-matrix–based propagation from the SALT2 parameter covariance to the distance-modulus covariance and then to the velocity covariance.

Changes:

• Replace the previous direct use of self._covariance_observation with a linear propagation via a design matrix A (and J) so that a 3N×3N SALT2 parameter covariance is mapped into an N×N distance-modulus covariance, including intrinsic scatter via sigma_M.
• Simplify the velocity covariance computation in give_data_and_variance to apply only the distance-modulus-to-velocity conversion matrix, given that the SALT2-to-distance-modulus propagation is now handled in compute_observed_distance_modulus_variance.
• Introduce self.number_datapoints to track N, use it consistently in shape checks and identity construction, and adjust _init_A and covariance-shape validation accordingly.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

_init_A initializes A with ones and then only ever assigns a subset of entries to 1 again, so the entire matrix stays filled with ones instead of having a single 1 per (SN, parameter) and zeros elsewhere. This makes the derived design matrix J effectively an all-ones matrix and causes the propagated distance-modulus covariance in compute_observed_distance_modulus_variance to be mathematically incorrect; A should be initialized to zeros and only the desired indices set to 1 so that each row selects the appropriate mb/x1/c element for that object.