Add opt-in setting to automatically include fixtures in migrations on change

[benjie]

Untested; just to demonstrate approach

[xvaara]

From claude opus (updated: #262 now builds on this PR, so my earlier comparison is superseded):

Short version: #262 is rebuilt on top of this PR. It now targets watch-fixtures and drops its own design. It adds the fixtures/auto/ folder from the Discord discussion, fixes a bug in the history lookup, and adds tests and docs. It's split into eight commits so you can review or cherry-pick them separately. The history fix is independent of the rest.

A bug here worth fixing on its own: in _reallyGetCommittedFixturesCache, the regex captures (^|\n) before the path, so const [, relativePath] = match reads that group. Every committed fixture is keyed by "" or "\n", none ever matches, and every fixture is included again in every migration. #262's first commit fixes it, and normalises ./x.sql to x.sql.

What #262 adds on top of this PR:

• Only fixtures/auto/ is included automatically, and creating the folder is the opt-in. The autoIncludeFixtures setting is removed. Without the folder, the committed history isn't read at all. Other fixtures, such as indexes, are only included manually.
• --!include inside fixtures/auto/ is an error. Nothing runs twice through nesting, and no DAG is needed.
• --! no-transaction inside fixtures/auto/ is an error too, as you suggested, instead of a warning that's easy to miss during commit. The automatic section always runs inside the migration's transaction, so such fixtures belong in another folder, included manually.
• The automatic fixtures go after the headers and --! no-transaction, and only that section is stripped on uncommit. With this PR alone:
  • commit loses --! Message: (the file is 000001.sql, not 000001-add-example.sql);
  • a current.sql that starts with --! no-transaction runs in a transaction;
  • the strip on uncommit removes everything before --!endAuto, including headers such as a -m message.
• Nothing can --!include an auto fixture, neither the current migration nor a fixture it includes, as you asked in review. Auto fixtures only ever run in the automatic section, so there's nothing already included to detect, and the TODO about include spellings goes away.
• New fixtures that are empty or only comments are skipped.
• Tests and docs: 19 tests (14 without a database, 5 lifecycle) and a README section. Before the setting was removed, I ran the 17 tests that existed then against this PR with it switched on, and 14 of them failed.

Corrections to my earlier comments:

• You were right that this PR lets you place a fixture by including it yourself; my earlier comparison made Auto-include fixtures from fixtures/auto, fix history lookup, add tests and docs #262 sound more flexible than it was. Following your review, Auto-include fixtures from fixtures/auto, fix history lookup, add tests and docs #262 now forbids that instead.
• Fix bug in commit error handling #275's approach of restoring exactly what the user wrote is better than Auto-include fixtures from fixtures/auto, fix history lookup, add tests and docs #262's uncompileIncludes for failed commits. Auto-include fixtures from fixtures/auto, fix history lookup, add tests and docs #262 no longer touches that.

Open questions:

• Projects that already have a fixtures/auto/ folder for something else would switch this on when they upgrade. Is a release note enough, or would a more distinctive folder name be better? At least any --!include of files in such a folder now fails with a clear error.
• One limitation remains: pg-minify 1.6.3 doesn't understand dollar quoting. Edits after -- inside a $tag$…$tag$ string go undetected, and some valid SQL (for example comment on function f() is $$Don't…$$;) makes it throw.