Retain more abundance information from S&T

[ethanplunkett]

Problem

Information is lost while preparing abundance to fit models.
During the Winter 2025 - 2026 Model QC we found it helpful to look at the ebird web visualization of the relative abundance to see two things that are not captured in the models:

1. The model coverage. These are no data in the eBird S&T rasters. During data prep we overwrite them with zero. See Edge and Island bias from cell aggregation  #12 and No data values in eBird abundance  #229.
2. The abundance prior to normalizing. In some weeks either due to poor detection or species exiting the modeled area ebird has much less abundance on the map. Normalization loses this information.

Proposed Changes

Store two more pieces of information in the BF model object:

• Model coverage ebird_model_coverage - Either a binary indicating the cell was in the modeled area or a percentage tof the constituent cells that were in the modeled area. This would have values for all cells in the extent as it needs to cover areas outside of the dynamic mask.
• Abundance prior to normalization We should store the total of the abundance for each week prior to the normalization. We could then add a new argument type to get_distr(). It would now have three options marginal , normalized (normalized S&T), raw (S&T prior to normalization). The default would be normalized matching the prior default. Deprecate the from_marginals argument BUT if from_marginals is set to TRUE force type to marginal so all existing code would be unaffected. raw is the new option and it would multiply the normalized distribution by the total for the week prior to normalization to restore the per-normalized version.