filter: Make under-sampling more apparent

[victorlin]

Context

Undersampling occurs in augur filter when the number of available sequences is lower than the targeted group size. This is not reported in any output. It is explained in this recently added docs section:

consider a dataset with 200 sequences available from 2023 and 100 sequences available from 2024. --group-by year --subsample-max-sequences 300 is equivalent to --group-by year --sequences-per-group 150. This will take 150 sequences from 2023 and all 100 sequences from 2024 for a total of 250 sequences, which is less than the target of 300.

Some historical context from #1454 (comment):

In the original formulation of only --sequences-per-group the idea was to say specify --sequences-per-group 10 and --group-by country would target 10 sequences per country and randomly sample these sequences for each country group. In the original formulation, we wouldn't top-up other countries. I think this is a semantic complication with adding the convenience parameter of --subsample-max-sequences. I'd think of --subsample-max-sequences as solely specifying --sequences-per-group.

Possible solutions

1. Add warnings. Example:

   WARNING: Targeted 150 sequences for group [year='2024'] but only 100 are available.
   

2. Add an option --output-group-by-sizes to highlight any discrepancies. Example:

   year	target size	available sequences	output size
   2023	150	200	150
   2024	150	100	100

Both solutions have been adopted for --group-by-weights in #1454, but they could be extended to other sampling methods.