Remove dead columns and tables left by deleted features #3015

Problem

Follow-up to the members.unsubscribed removal (#3007, merged in #3013/#3014). A scan of every table and column in the schema (46 tables, 377 columns) found four orphaned tables and three dead columns. Each item's feature code was removed months or years ago; only the schema and the production rows remain.

Findings

Dead tables — no model, controller, route, or view references remain:

Dead columns on live tables:

events.coach_description has one residual write: spec/fabricators/event_fabricator.rb sets a Faker value. It goes with the drop.

Work plan

Sub-issues, ordered. Each column drop follows the #3007 two-step (see "Why the ignore step comes first"): a PR that adds the column to ignored_columns, then a separate PR that drops it.

1. Remove the stale ignored_columns entries from Member — both referenced columns are already gone.
2. Drop meetings.invites_sent (two PRs).
3. Drop workshops.sign_up_url (two PRs).
4. Drop events.coach_description (two PRs; includes the fabricator cleanup).
5. Drop the four dead tables (one PR; no ignore step needed because nothing references them).

Why the ignore step comes first

Same reasoning as #3007, shortened: the release phase runs migrations before new dynos boot, and under Heroku preboot old dynos keep serving against the new schema during the window. Rails 8.1 builds explicit column lists, so an old release serving after the drop would fail every query on that table with PG::UndefinedColumn. Adding the column to ignored_columns and deploying first takes it out of the new release's SELECT list; the second deploy then drops it safely.