Exercise 2: Adding a Required Column to a Live Table — Possible Solution
==========================================================================

Why not just add it?
  "excerpt String" on a table with rows fails: existing rows would have
  no value. A default like @default("") avoids the error but leaves
  every existing post with an empty excerpt, which is meaningless.

Plan (each step is its own deploy):

Deploy 1 — expand
  schema.prisma:  excerpt String?
  $ npx prisma migrate dev --name add-post-excerpt
  SQL: ALTER TABLE "Post" ADD COLUMN "excerpt" TEXT;
  Code: unchanged. Old and new code both still work.

Deploy 2 — write it
  Code: createPost and updates now always set excerpt, e.g. the first
  200 characters of the content, trimmed at a word boundary.

Deploy 3 — backfill old rows (data migration)
  $ npx prisma migrate dev --name backfill-post-excerpt --create-only
  migration.sql:
    UPDATE "Post"
    SET "excerpt" = LEFT(COALESCE("content", "title"), 200)
    WHERE "excerpt" IS NULL;
  (For a very large table, run the same UPDATE in batches from a
  script instead, e.g. by id range, to avoid long locks.)

Deploy 4 — make it required
  schema.prisma:  excerpt String
  $ npx prisma migrate dev --name require-post-excerpt
  SQL: ALTER TABLE "Post" ALTER COLUMN "excerpt" SET NOT NULL;
  This succeeds because no NULLs remain, and running code always
  writes a value since Deploy 2.

Verify before Deploy 4:
  SELECT COUNT(*) FROM "Post" WHERE "excerpt" IS NULL;   -- must be 0

WHY THIS WORKS AS AN ANSWER
------------------------------
At every step, the database accepts what both the currently running
code and the next version do: the column starts optional, new code
fills it in, the backfill covers old rows with real content, and only
then does the NOT NULL constraint arrive. The backfill is re-runnable
(WHERE "excerpt" IS NULL), and the check before the final step turns
"I think it's done" into a measured fact.