scaffolding regression fixes plus docker updates

[dpark01]

This PR:

1. fixes a bug [bug] assemble_denovo fails at larger ANI distance #543 introduced recently that causes different/unintended failure modes in assemble_denovo on example exercises. This PR now has the scaffold task fallback to old brute-force reference selection if ANI-based reference selection fails to find any matches at all. This will work fine (same as before) in most historically-normal use cases, but will not behave well if given a very large array of reference genomes to choose from. (this change does not impact the behavior of scaffold_and_refine_multitaxa which does not pass multiple references to the scaffold task anyway.
2. Updates viral-core and viral-assemble docker images to latest (2.3.2 and 2.3.2.0)
3. Updates cromwell, womtool, and dxCompiler versions for build
4. minor cleanups to docbuild
5. drops any direct use of util.file.zstd_open in favor of zstandard.open
6. increase default RAM for reports.alignment_metrics
7. bugfix some bed file sorting for samtools ampliconstats in the case of multi-segment targets -- also allow samtools ampliconstats to fail silently in the case that it's getting too picky about the bed file
8. change the mafft workflow to use multi_align_mafft_ref instead of multi_align_mafft
9. make the terra_tsv_to_table workflow resilient to non-existent TSVs in its input array, to simplify running it on a Terra table full of tsvs that may or may not exist
10. Some initial WiP code for multi-species genbank prep, not far along yet.

[tomkinsc]

Lines 323-325 are bit repetitive with line 305 above, and it may be nice to also have these values in the log for debugging—what about something like:

chosen_ref_data_fields=($(sed -n '2{p;q;}' "~{sample_name}.refs_skani_dist.full.tsv"))

CHOSEN_REF_FASTA="${chosen_ref_data_fields[0]}"
tee <<< "${CHOSEN_REF_FASTA}" CHOSEN_REF
tee <<< "${CHOSEN_REF_FASTA%.*}" CHOSEN_REF_BASENAME

tee <<< "${chosen_ref_data_fields[2]}" SKANI_ANI
tee <<< "${chosen_ref_data_fields[3]}" SKANI_REF_AF
tee <<< "${chosen_ref_data_fields[4]}" SKANI_CONTIGS_AF

[tomkinsc]

The merge of this to master triggered a deployment to dnanexus, which is currently failing to compile via dxWDL:
https://github.com/broadinstitute/viral-pipelines/actions/runs/10237505135/job/28320935779
The issue seems to be with tasks_megablast.wdl. The checks in this PR didn't catch it because deployment to dx is currently only triggered after a merge into master. I'm looking into it and may change actions behavior so we actually test deployment to dx in PRs.