Big Sam job reliability fixes, dry-run, and email reports

[jayvarner]

Summary

Brings main up to date with develop - main is currently missing all of this work, which is why the deployed ECS task is crash-looping (the SES boot-order fix included here is what actually resolves that).

• Elasticsearch 8 refactor and related test/DB config fixes
• LoadBigSamJob reliability pass: per-row transactions with rollback on failure, structured error/skip reporting instead of silently swallowed rescues, fixed several confirmed bugs found by auditing against the real production spreadsheet (case-sensitive flag columns, duplicate repository/collection labels, a copy-paste bug wiring the wrong collection to the third repository slot, a dead notes-field typo, a NoMethodError that could crash the whole job on certain recipient names, and blank spreadsheet IDs silently colliding into one letter record)
• LoadBigSamJob#dry_run - runs the full pipeline and reports what would happen without persisting anything or touching Elasticsearch
• Decomposed the 234-line, cyclomatic-complexity-78 process_letter into focused methods; the three near-duplicate repository/collection blocks collapsed into one table-driven method
• SES email reports after every real load: a developer report and a plain-language owner report, sent via BigSamMailer
• Fixes the ActionMailer/SES boot-order bug (ses_settings= NoMethodError) currently crashing the ECS task

Test plan

• Full relevant spec suite green on develop (job, mailer, model specs - 50 examples)
• Verified the SES fix directly by booting with RAILS_ENV=production and confirming ActionMailer::Base.delivery_method/ses_settings resolve without error
• Redeploy ECS from main after merge and confirm the task boots past db:migrate